mirror of
https://github.com/coder/coder.git
synced 2026-11-20 09:58:59 +08:00
feea5ab959f59cd2135443ed2a0c19bf7643db8f
5064
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
|
|
feea5ab959 |
feat: add GPT-6.2 Astra and GPT-6.2 Sol to Agents and AI Gateway (#32370)
Closes CODAGT-1608 This adds `gpt-6.2-astra` and `gpt-6.2-sol` to the Agents known models catalog, so they're suggested when an admin adds a chat model, and regenerates the AI Gateway price book from models.dev so the gateway can price them. | Model | Input | Output | Cache read | Cache write | |---|---|---|---|---| | GPT-6.2 Astra | $8 | $40 | $0.40 | $10 | | GPT-6.2 Sol | $2 | $10 | $0.10 | $2.50 | Prices are per 1M tokens, and both models have a 1,050,000 token context window and 128,000 max output tokens. They're in the catalog for OpenAI (with a suggested effort of `medium`, same as `gpt-6-astra` and `gpt-6.1-sol`), OpenRouter and Vercel. models.dev doesn't list them on Azure or Bedrock yet, so I've left those out for now. Both models also reject the `none` and `minimal` reasoning efforts. A model config only sets a default and a maximum effort, so the picker always offers both, and picking either one fails the turn: ``` Unsupported value: 'none' is not supported with the 'gpt-6.2-astra' model. Supported values are: 'low', 'medium', 'high', 'xhigh', and 'max'. ``` We already send `low` instead for `gpt-6-astra` and `gpt-6.1-sol`, so to fix, we'll just do the same for the two new models. All four now live in one list of model ID prefixes, rather than a helper each. I checked this against the real models: all four return a 400 like the one above for `none` and `minimal`, and accept `low`. `gpt-6-sol` and `gpt-6.1-terra` still accept `none`, so they're left alone. Regenerating the price book also brings in everything else models.dev changed since the last weekly refresh, mostly on OpenRouter and Vercel: 22 other new entries (4 of them `-pro` and `-fast` variants of these two), 29 it dropped and 51 price changes. 6 existing catalog entries get new prices or limits the same way. |
||
|
|
7c7e2e73cf |
feat: honor disable-model-invocation in uploaded skill content (#31159)
Addresses @ethanndickson's [review comment on #30882](https://github.com/coder/coder/pull/30882#discussion_r4344815088): the same `SKILL.md` file could be manual-only in Claude Code and the project dialog but automatic in Coder, because the server and the browser each handled `disable-model-invocation` in their own way. The server now owns the key for personal, organization, and project skills: - When a create or update request leaves `model_invocable` unset, `disable-model-invocation: true` in the content turns automatic use off, and `false` turns it back on. The key is read with the function workspace skills use, so a file gets the same setting in both places. Only YAML booleans count, so a quoted `"true"` is ignored. - An explicit `model_invocable` in the request always wins. - New content without the key leaves the setting unchanged, so editing text never turns it back on. - Content is stored as uploaded. Every API response returns content whose key agrees with `model_invocable`: when the stored file says otherwise, the server replaces the value of the key or adds the key before the first one, and leaves every other byte alone. - **Download** and **Export all** save the content the API returns, so the browser no longer edits the file. `withDisabledModelInvocation` is removed. - An upload is rejected with a 400 when the key holds a value the server can't replace with a boolean, such as a list. A later change to `model_invocable` can then always reach the file. Two edges: - Adding the key makes the returned file about 30 bytes larger, so a file at the 64 KB upload limit can be too large to upload again. - A skill stored before the upload check with a value the server can't replace is returned as stored. Workspace skills already honor the key since #31147. #31160, the frontend PR on top of this one, adds the switch to the personal and organization skill editors. > Generated by an agent on behalf of the user --------- Co-authored-by: Cursor <cursoragent@cursor.com> Co-authored-by: Nick Vigilante <nickvigilante@users.noreply.github.com> |
||
|
|
da0bd0750f |
chore: remove most export_test.go shims (#32162)
Follow-up to https://github.com/coder/coder/pull/25601, where I missed the other way agents avoid `*_internal_test.go`: an `export_test.go` file that re-exports unexported symbols, just so a `package foo_test` test can reach them. There were 13 of these on `main`, most recently from #32063. This PR removes 8 of them. Tests of package internals are now `*_internal_test.go` files, with no changes beyond the package name and the symbols they call. `messagepartbuffer` and `agentcontext` only needed one small accessor each, so `Buffer.EpisodeCount` and `Manager.Started` are just exported for real, and their tests stay where they were. To stop new ones appearing, the `testpackage` linter no longer skips `export_test.go` (the 5 that are left are excluded by path), and `AGENTS.md` and `.claude/docs/TESTING.md` now state the rule. `export_test.go` is an agent idiom, and imo we should try and get rid of it. So, once this PR is approved, we'll fix up all the remaining ones in a follow-up PR, somehow, and undo the lint exclusion when we do. |
||
|
|
166b9d25e3 |
test(coderd): remove low-value cron and chatwebhook tests (#32164)
The tests from #32063 and #32062 are two to three times the size of the code they cover, and a lot of them can't fail for a reason we'd care about. This PR removes those, with no production changes. In `coderd/schedule/cron`, two tests rebuilt the schedule rule a second time to compare against, which was also the only use of `export_test.go`, so that goes too. Most of the rest, in both packages, are table rows that take the same branch as a row that stays. I also dropped a copy of the webhook scheme list and a repeated-request test that belongs at the route. What's left is a row per reject branch, the RFC cases, the regression test for the `Next` hang and the constant-time scan. I replayed each fix commit from the two PRs against the smaller suite, and the pre-fix code still fails a test for every one, other than a fix to the wording of a zone error. One row that stays is changed: the bearer wrong-scheme case used `Basic`, which was only rejected because of where its space lands, so nothing covered the scheme word. It's now `Digest`. #32162 overlaps with this in `cron`. With the shim's users gone it no longer needs to touch that package, so it's now stacked on top of this PR. |
||
|
|
a3f25a491e |
feat: forward bridged routes in AI Gateway proxy mode (#30226)
Enable bridged request forwarding in AI Gateway proxy mode. Responses are captured through `TeeBuffer`. Proxies are created per-request reusing transports. Handle `http.ErrAbortHandler` inside the circuit-breaker wrapper so aborted responses are classified by HTTP status before the panic is propagated. Generated by Coder Agents on behalf of @pawbana. |
||
|
|
fe0b6e5de0 |
fix: take chat state from the stream only (#30478)
The chat page got a chat's messages, queue and status from three places: REST pages, action responses and stream events. None of them had a version, so the client could not tell which was newer and reconciled them with rules of its own. A refetch or a response could replace newer state from the stream, and a stream message that failed to parse lost its events and left an error until a reload. The server now returns snapshot_version on the newest messages page and on stream events (#29662), which makes one copy enough. Each chat has a store that loads the newest page once; after that, only stream events and older pages change it. To reconnect, to return to a chat within five minutes, or to recover from a failed parse, the client opens the stream again at the highest version it applied. It refetches no messages, so no response replaces what the stream set. The client no longer reads messages or queue rows from action responses, so a send, an edit or a queue change would show only once the stream reports it. The store therefore shows the action's direct effect at the click and keeps it until the stored state holds the result or the request fails. It does not end at the next stream event of a kind, because that event can report another change, and the user's change would disappear and come back. A sent message has no ID until the server creates it, so pending sends are matched to new messages in click order, and a chat's send, edit and queue requests run one at a time to keep that order. This assumes one user in one client. A message from an agent, an automation or another API client can end a pending send early, and the next stream event corrects it. The composer clears at the click, because the chat already shows the sent message. It stays editable, so text typed during the request is kept. Closes CODAGT-903 |
||
|
|
99e3bf91a7 |
fix(coderd/database/migrations): resolve duplicate version (#32329)
Main cannot initialize database migrations because #32309 reused version 000643, already taken by the skill display-name migration from #31747. Renumber the later chat-project migration and its fixture to 000644, preserving the earlier migration number. All three renames leave SQL contents unchanged. |
||
|
|
3211ca066e |
feat: add organization unpriced AI models endpoint (#31753)
## Description
Adds the experimental `GET
/api/experimental/organizations/{organization}/ai/model-prices/unpriced`
endpoint behind the "N models not set up" warning on the organization AI
spend page
([Figma](https://www.figma.com/design/6TkixpkUIM5Qv6I7qhM3xZ/AI-Gov---user-spend?node-id=70-11964&t=KQwENGjcQvVT9qkh-4)).
It lists the models used in the period that still hold no price in the
organization, using the query from this stack.
## Changes
- Add the `organizationUnpricedAIModels` handler and
`ExperimentalClient.OrganizationUnpricedAIModels`:
- Reuses `aiSpendRetainedPeriod`, since token usage is subject to data
retention.
- Returns the applied period, `count`, and `models` (provider name,
display name, icon and type, model, token count), most unpriced tokens
first.
- Requires read on the organization's model prices (owners and
organization admins), so the UI only requests it for users who can set
prices.
- Add handler tests for the response, period validation, organization
scoping and role access.
Note: pagination will be added in a follow-up.
Closes AIGOV-730
> [!NOTE]
> Generated by Coder Agents on behalf of @ssncferreira.
|
||
|
|
1f1aa6ae3c |
feat: add chat project template schema and queries (#32309)
Adds the database layer for attaching workspace templates to chat projects: the schema, the queries, their authorization, and the audit type. Nothing calls the new queries yet, and the existing create and update handlers pass the new parameters unset. The API, the limit's enforcement, and the agent tools follow in the PRs above. - Attachments live in a `chat_project_templates` table (`project_id`, `template_id`, `created_by`, `created_at`), deleted with the project or the template. `chat_projects` gains `limit_templates` (default `false`), `default_template_id`, `default_template_version_preset_id`, and `default_parameters jsonb`. A project stores a default preset or default parameter values, not both. - `InsertChatProjectTemplates` needs project `update` and `use` on every template. It skips templates that are already attached, and templates outside the project's organization, so an attachment never crosses organizations. `DeleteChatProjectTemplate` needs project `update` and template `read`, and clears the project's default when it detaches the default template. Both checks are in dbauthz. - `GetChatProjectTemplatesByProjectIDs` is authorized through the project, and leaves out deleted, deprecated, and agent-disabled templates. It does not check the caller's access to each template, so callers do that before they expose one. - `UpdateChatProjectByID` changes the template settings only when the request sets them, so an update that omits them cannot write back a stale snapshot. - `AuditableChatProjectTemplates` is the audit form of a project's attached templates, logged as a write to the project. The new `chat_projects` columns are tracked in the project's own audit diff, with the default parameter values masked because they may hold credentials. **NOTE**: the migration adds one table and four columns and changes no existing rows. Existing projects keep unlimited template access. Stack (merge top to bottom): - #32309 👈 - #30754 - #32310 - #30755 - #30756 - #30898 |
||
|
|
c3ff770e8a |
fix(coderd/database): match workspace requester by owner ID when sorting (#32188)
GetWorkspaces sorted the requester's favorite and own workspaces first by matching the owner's username. Usernames are unique only among non-deleted users, and a deleted user keeps theirs, so a user who took a deleted user's username saw that user's deleted workspaces and favorites sorted as their own when listing deleted workspaces. Match by owner_id instead, as the sort did before #22290. The username form most likely worked around the sqlc CTE bug: with the EXISTS (tasks) filter the query had then, sqlc without the coder/sqlc fix rejected the bare owner_id as ambiguous. #29085 removed that filter, so the owner_id form generates with either sqlc. |
||
|
|
4866d0fb1f |
feat: add display names to project skill actions (#31747)
closes CODAGT-1506 Project skill action chips show the skill's slug (`deploy-bot`, `security-review`), because skill names must be kebab-case. This adds an optional display name that labels the chip instead. Project skills get a `display_name` column, set through `display_name` on the create and update requests. It is limited to project skills (the same rule as `action`), trimmed, single line, and at most 64 Unicode code points. The chip, its tooltip, and the "More" menu show the display name, sort by it, and fall back to the skill name when it is empty. When a display name is set, the tooltip and the menu item also show the `/name` slash command, so actions that share a display name stay distinguishable. Clicking an action still inserts `/name`, and the `/` menu is unchanged. The project skill editor shows a **Display name** field under **Show as action** only while that switch is on. Turning the switch off hides the field but keeps its value, and saves omit `display_name` while the action is off, so the stored value stays. The display name lives in the database rather than SKILL.md frontmatter, so downloading and re-importing a skill does not carry it. To share the code-point length check with the editor, `countCharacters` and `maxCharacters` move from `ChatProjectDialog.tsx` into `site/src/utils/formUtils.ts`. <img width="637" alt="Project skill editor with the Display name field under Show as action" src="https://raw.githubusercontent.com/coder/coder/assets/pr-skill-action-display-name/pr/edit.png" /> <img width="520" alt="Action chips labeled with display names, with the tooltip open" src="https://raw.githubusercontent.com/coder/coder/assets/pr-skill-action-display-name/pr/chips.png" /> Refs CODAGT-1506 |
||
|
|
3795d72713 |
fix: search AI Gateway models and clients by substring (#32159)
Closes AIGOV-801 Typing `hai` into the model filter on the AI Gateway sessions page shows the haiku models for a moment, then "No models found". Whilst the debounced request is pending, the menu filters the options it already has by a case-insensitive substring, but `GET /api/v2/ai-gateway/models` only matched a case-sensitive prefix, and Bedrock identifiers start with `anthropic.`. The client filter matched the same way, so it had the same problem. To fix, both endpoints now match a case-insensitive substring. This needed one change to the model menu. When the page loads with a model already in the filter, the menu used to look that model up on the server, by searching for its name and taking the first result. That's the same pattern as the user filter, where the lookup fetches the avatar, but a model option needs nothing from the server, so all it did was check that the model exists. With a substring search, the first result for `claude-3-5-sonnet-20241022` can be `anthropic.claude-3-5-sonnet-20241022-v2:0`, so the check would fail for a model that does exist. The menu now skips the lookup and just shows the model from the filter as selected. The only visible difference is that a model name in the URL that doesn't exist now shows as selected, instead of "All models". The client menu has the same lookup, which still works, as no client name contains another, so I've left it alone. The prefix match was added in #22230 (models) and #22694 (clients) so the queries could use the B-tree indexes. That only works on a C collation database, and only for models, as the clients query wraps the column in `COALESCE`. On an `en_US.utf8` database it's a plain filter either way. On a C collation database a model search with no matches does get slower, from instant to a few seconds with 5M synthetic rows, but I think that's worth it for a search that works. |
||
|
|
0be6e7a034 | feat: add a setting for web terminal font size (#32027) | ||
|
|
b0ae8adab5 |
feat(coderd): protect pending jobs only while the provisioner queue is moving (#32099)
Stops the pending job reaper from failing jobs in a queue that's still moving. This is an alternative to #31469 that came out of Steven's [review](https://github.com/coder/coder/pull/31469#pullrequestreview-5847958482) asking whether 2h is enough, and it builds on George's work there. A pending job past 30m is now only protected while a matching online provisioner is making progress, meaning it finished a job in the last 30m or is running one that's still being updated. A moving queue never gets reaped and a frozen one still does at 30m like today. There's no absolute cap. No migration since it uses the existing `provisioner_jobs_worker_id_organization_id_completed_at_idx`. One tradeoff is that a stuck build whose provisioner keeps heartbeating counts as progress, so if every matching provisioner is stuck like that, the queue waits until that build is canceled. Still a draft so we can talk it through with Spike and Steven. Part of CORE-586 --------- Co-authored-by: George Katsitadze <george.katsitadze@gmail.com> Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com> |
||
|
|
d412c38afb |
perf: validate chat transitions from the locked row (#30886)
One of the main costs in terms of DB load from chats is extra round trips during transactions. This results in locks being held for longer periods of time -> an increase in lock waits for chats related queries. One of the opportunities for reducing some of the round trips is to eliminate extra chat row reads after we have just read the row in order to lock it. In this PR `ChatMachine.Update` now seeds the transaction struct with the row returned by `LockChatAndBumpSnapshotVersion`, plus a queue count taken right after it. We can then pass it into `Update` as an optional parameter validate against it, so neither re-reads the chat row from the database while the row lock is held. `updateChat` (send, edit, promote, tool results, interrupt, compact, clear, reconcile), hook failure, family archive, and the worker's generation and task transitions use it instead of reloading the chat under the lock. This PR relative to main: |measurement | main | this PR | percent change | |-- | -- | -- | -- | | Round trips under lock | 180.9 | 142.0 | -39% | | Statements under lock | 86.1 | 66.6 | -23% | | GetChatByID under lock | 54.1 | 35.8 | -34% | | GetChatByID total | 75.5 | 57.1 | -25% | | Statements | 235.8 | 216.2 | -8% | Generated by Coder Agents on behalf of @cstyan. |
||
|
|
54fe847c6f |
chore: remove references to torn-down dev.coder.com instance (#31702)
The `dev.coder.com` instance has been torn down and now only redirects to another instance. Nothing in the repo calls it. Every reference was a test fixture, mock, story, Helm test input, docs example, or comment. Replace all occurrences with `coder.example.com` and update the comments that claimed the mocks were fetched from that instance. The Helm golden files were updated to match their inputs. Closes DOG-43 https://linear.app/codercom/issue/DOG-43/scan-codercoder-repo-for-remaining-references-to-devcodercom-and |
||
|
|
4f6a56427e |
feat: report pubsub health from the pubsub backend (#31621)
Fills in the `pubsub` section of the deployment health report on the server. Stacked on #31620. - New `pubsub.ReportingPubsub` interface: `Pubsub` plus `HealthReport(ctx) healthsdk.PubsubReport`. The `Pubsub` interface itself is unchanged. - The PostgreSQL, NATS and in-memory backends implement it. Each reports its backend name (`postgres`, `nats` or `memory`) and its connection state: - **PostgreSQL:** tracks the `pq` listener's connected state from its event callback. When it's not connected, the report has severity `warning` and `EPS01`. - **NATS:** warns with `EPS02` if the embedded server isn't `Running()`, and with `EPS01` for each publish or subscribe connection whose `Status()` isn't `CONNECTED`. - **In-memory:** always `ok`. - `coderd.Options.Pubsub` is now a `ReportingPubsub`. `ReplicaSyncPubsub` stays a plain `Pubsub`. - `coderd/healthcheck` has a new `Pubsub` check, which counts toward `healthy` and the overall severity like the other sections. `formatHealthcheck` applies the dismissed setting for it. - New health codes `EPS01` (not connected) and `EPS02` (embedded NATS server not running), documented in `docs/admin/monitoring/health-check.md`. - Test helpers follow the type change: `coderdtest.Options.Pubsub` and the pubsub returned by `dbtestutil.NewDB` are now `ReportingPubsub`, and a few test wrappers embed `ReportingPubsub` instead of `Pubsub`. <details> <summary>Decision log</summary> - `HealthReport` returns a `healthsdk.PubsubReport` rather than filling one in place. This matches the `healthcheck.Checker` methods. `ctx` is unused for now. - The healthcheck layer sets `Dismissed`, so backends don't need to know about health settings. - Problems are reported as warnings, not errors, so they show on the health page without marking the deployment unhealthy. Both backends reconnect on their own. - Postgres connection state comes from the listener event callback, not from `pq.Listener.Ping`, so the health check doesn't do database I/O. Closing the pubsub also marks it disconnected. - NATS `conn` gains `Status()`, which `*natsgo.Conn` already implements. The pools are immutable after construction, so `HealthReport` reads them without locking. - The docs say the pubsub section appears in the JSON response, because the health UI doesn't show it until a later PR in this stack. - psmock is not regenerated: none of its users pass it to coderd. </details> --- Generated by Coder Agents on behalf of @spikecurtis. |
||
|
|
2187ecd37e |
feat: add PubsubReport to healthsdk HealthcheckReport (#31620)
Adds pubsub to the deployment health report types in `healthsdk`:
- New top-level `pubsub` section (`PubsubReport`) in
`HealthcheckReport`. It embeds `BaseReport` and adds one field,
`backend` (string).
- New `HealthSectionPubsub` constant, added to `HealthSections` so the
section can be dismissed.
- `HealthcheckReport.Summarize` includes pubsub errors and warnings,
which `coder support bundle` prints.
This is the first PR in a stack for GRU-100. It only changes the SDK and
the files generated from it (Swagger, API docs, `typesGenerated.ts`),
plus the two frontend health mocks that need the new required field. The
server does not fill in this section yet. That comes in a follow-up PR.
<details>
<summary>Plan</summary>
- Add `PubsubReport { BaseReport; Backend string }` to
`codersdk/healthsdk`.
- Add `Pubsub PubsubReport \`json:"pubsub"\`` to `HealthcheckReport`.
- Add `HealthSectionPubsub` and include it in `HealthSections`.
- Include `Pubsub` in `HealthcheckReport.Summarize`, with test coverage.
- Regenerate TS types, Swagger and API docs.
- Follow-up PRs: populate the report in `coderd/healthcheck` (including
severity and dismissal handling) and update the health UI.
</details>
---
Generated by Coder Agents on behalf of @spikecurtis.
|
||
|
|
38e74c4506 |
feat: add agent-plugins experiment and Agent API v2.17 plugin messages (#29320)
Linear: [CODAGT-1101](https://linear.app/codercom/issue/CODAGT-1101/agent-plugins-read-only-support-behind-the-agent-plugins-experiment) **tl;dr:** Adds the `agent-plugins` experiment and the Agent API v2.17 messages that carry Agent Plugins data from the agent to coderd. First of eight stacked PRs adding read-only [Agent Plugins](https://agent-plugins.org/) v1.0.0 support. A plugin is a directory with a `plugin.json` manifest, skills under `skills/`, and an optional `mcp.json`. Across the stack, the agent discovers plugins in `plugins/` and `.agents/plugins/` under its scan roots (for example `~/.coder/plugins/<name>`), chatd lists their skills with plugin attribution, the agent runs the MCP servers of plugins in the workspace working directory, and the Agents page shows plugin attribution in the chat context. Hooks, commands, agents, install flows, and org-managed plugins are out of scope. Background: [research/agent-plugins-standard.md](https://github.com/coder/scott-misc/blob/main/research/agent-plugins-standard.md) (internal). This PR adds `codersdk.ExperimentAgentPlugins` and a `PluginBody` oneof variant on `ContextResource` (tag 18) that carries the validated manifest fields and nests the plugin's skills and MCP servers. `Manifest` gains `agent_plugins_enabled` (field 23), which coderd sets when the experiment is enabled; the agent discovers plugins only then. coderd stores each nested resource as its own row with the plugin's name in a new `plugin_name` column (migration 000642), which the hydrate, MCP sync, and re-pin queries copy. With the experiment off, `PushContextState` drops plugin resources and everything nested in them. Otherwise it rejects pushes with plugin names that fail `workspacesdk.ValidatePluginName` (the spec's name grammar, shared by agent and coderd), OK plugins without a name or with a duplicate name, a non-OK plugin that nests resources, nested entries of the wrong kind, plugin skill sources outside `<plugin source>/skills/`, and plugin MCP servers with an empty `server_name` or a source other than `<plugin name>/<server_name>`. Nested resources count toward the 1000-resource cap. Agents now dial v2.17, so a new agent needs a coderd that supports it, as with every Agent API minor. Turning the experiment off does not remove plugin data already pinned to existing chats. <!-- pr-map:start --> <details><summary>PR map</summary> `main..4b0cfc9657` | Category | Files | + | - | |---|---:|---:|---:| | prod (backend) | 15 | 342 | 78 | | test (backend) | 9 | 534 | 3 | | migrations | 2 | 14 | 0 | | generated (backend) | 7 | 1216 | 1027 | | generated (frontend) | 1 | 2 | 0 | | **Total** | 34 | 2108 | 1108 | 1. `coderd/agentapi/context.go` (+211/-51, API): drops plugin resources when the experiment is off; otherwise validates plugin names and nesting, and stores each nested skill and MCP server as its own row with the plugin's name. 2. `agent/proto/agent.proto` (+39/-9, API): `PluginBody` with nested `skills` and `mcp_servers`, and `Manifest.agent_plugins_enabled`. 3. `codersdk/workspacesdk/frontmatter.go` (+31/-0): `ValidatePluginName` and `MaxPluginNameLength` (manifest name grammar). 4. `coderd/database/queries/chats.sql`, `workspaceagentcontext.sql`, migration 000642 (+28/-7, DB): the `plugin_name` column on both context resource tables, copied by the hydrate, MCP sync, and re-pin queries. 5. `codersdk/agentsdk/agentsdk.go`, `agent/agent.go` (+18/-4, API): `Manifest.AgentPluginsEnabled` and `ConnectRPC217WithRole`; the agent dials v2.17. 6. `coderd/agentapi/api.go`, `manifest.go` (+4): the manifest flag follows the experiment. 7. `tailnet/proto/version.go` (+8/-1, API): Agent API v2.17. 8. `codersdk/deployment.go` (+4, API): the `agent-plugins` experiment. </details> <!-- pr-map:end --> --- This PR was generated by Coder Agents on behalf of @jscottmiller. Co-authored-by: Garrett Delfosse <delfossegarrett@gmail.com> |
||
|
|
d611c60a9a |
feat: scope unpriced AI models to an organization and its prices (#31752)
## Description Scopes `GetUnpricedAIModels` to an optional organization and checks organization prices, so the unpriced models endpoint can list what an organization still needs to price. Usage is attributed to an organization through its token usage's effective group, the same as spend. ## Changes - Optional `organization_id`: when set, only that organization's usage counts, and when null (the notification), all usage counts. - A model still holds a price if it has a site-wide price for the provider type, or an organization price (`ai_model_configs`, live and with a price) for the provider in the usage's organization. - dbauthz reads `ResourceAiModelPrice`, scoped to the organization when one is given, like `GetAIModelPriceForProvider`. - Table-driven query tests run every case for all organizations, org A and org B. <details> <summary>Cases</summary> Model X is on `anthropic-prod`. Y is the model the provider reports serving for X. "Notif" is the deployment-wide notification, "UI A/B" is the endpoint for org A/B. | # | Prices | Usage | Notif before | Notif after | UI A | UI B | |---|---|---|---|---|---|---| | 1 | No price for X anywhere | A uses X, unpriced | ✅ | ✅ | ✅ | ❌ (B didn't use X) | | 2 | Site-wide price for X | A and B use X, priced | ❌ | ❌ | ❌ | ❌ | | 3 | A priced X before the usage | Only A uses X, priced | ❌ | ❌ | ❌ | ❌ | | 4 | A priced X after the usage | Only A uses X, unpriced when recorded | ✅ (wrong) | ❌ | ❌ | ❌ | | 5 | A priced X, B has none | A uses X (priced), B uses X (unpriced) | ✅ (because of B) | ✅ (because of B) | ❌ | ✅ | | 6a | No price for X or Y when recorded, then Y priced site-wide | Only A uses X, unpriced when recorded | ✅ | ✅ | ✅ | ❌ | | 6b | Same as 6a | A and B use X, unpriced when recorded | ✅ | ✅ | ✅ | ✅ | | 7a | No price for X, but a site-wide price for Y | A uses X, priced through Y | ❌ | ❌ | ❌ | ❌ | | 7b | No price for X, A priced Y, and B has none | A and B use X: A priced through Y, B unpriced | ✅ (because of B) | ✅ (because of B) | ❌ | ✅ | </details> Note: pagination will be added in a follow-up. Related to AIGOV-730 > [!NOTE] > Generated by Coder Agents on behalf of @ssncferreira. |
||
|
|
8c5921db5e |
fix: count unpriced AI model usage by when it was recorded (#31751)
## Description Updates the unpriced AI models query behind the weekly owner notification to count token usage by when it was recorded (`token_usages.created_at`) in a `[period_start, period_end)` window, consistent with the other AI spend queries. It also returns one row per provider name and model instead of per provider type, so the unpriced models endpoint later in this stack can reuse it. The notification payload keeps the same format. ## Changes - Rename `GetUnpricedAIModelsSince` to `GetUnpricedAIModels`. It used `interceptions.started_at` with no end. - Treat usage as unpriced only when no price was found (`cost_micros IS NULL AND priced_model IS NULL`), so usage whose price was found but whose cost could not be computed is not reported. - Return one row per live provider name and model, with the provider's display name, icon and type. - The notification passes the last 7 days and merges rows by provider type, since site-wide prices are keyed by type, so its payload keeps the same shape. - Table-driven query tests and a notification test for providers sharing a type. Related to AIGOV-730 > [!NOTE] > Generated by Coder Agents on behalf of @ssncferreira. |
||
|
|
5c9d74e019 |
perf(coderd/x/gitsync): poll merged PRs hourly (#31922)
Merged PRs cannot reopen, so their diff-status rows do not need the two-minute TTL used for open and closed PRs. Use a one-hour TTL after successful diff-status refreshes of merged PRs while retaining periodic metadata updates. Workspace git reports still set `stale_at` to the past so merged diff-status rows remain eligible for the next worker tick. Refs CODAGT-1543 |
||
|
|
d639ced1e0 |
fix: send only changed chat history over the stream (#29662)
A client could resume the chat stream only with after_id, a message ID. A message ID does not identify what a client has received. A reconnect could skip a message that another client sent, send no history_reset for an edit the client missed, or resend the whole history. A client can now resume with the chat's snapshot_version, which the newest messages page returns and the stream's events then include. The stream sends what changed at that version or later. For an edit it sends the ID of the edited message and the messages written since, not the whole history. A stream opened without snapshot_version sends the same events as before. A write outside a chat state transition does not increase snapshot_version, and a stream can miss it. The queries that only tests used for such writes are removed. When the stream cannot be set up, the server now returns 500 before the WebSocket upgrade. It used to upgrade and send an error event, which the web client shows as a chat error. The Tasks to Chats migration guide told clients to reconnect with after_id. It now describes snapshot_version. |
||
|
|
30abd0593c |
feat: add organization AI spend overview endpoint (#31709)
## Description
Adds the experimental `GET
/api/experimental/organizations/{organization}/ai/spend` endpoint that
backs the new Overview tab of the organization AI spend page ([Figma
design](https://www.figma.com/design/6TkixpkUIM5Qv6I7qhM3xZ/AI-Gov---user-spend?node-id=12-5152&t=YlCC7RGSMNrzMxVo-4)).
It returns the organization's spend summary, daily spend, and a page of
users with their per-group spend, using the queries from #31711.
## Changes
- Add the `organizationAISpend` handler and
`ExperimentalClient.OrganizationAISpend`:
- Reuses the period handling of `/ai/spend/users`: `period_start` and
`period_end` must be provided together and span at most 31 days, and the
default is the current budget period.
- Also requires both bounds at 00:00:00 UTC, since the daily spend
aggregate holds whole UTC days.
- Is not limited by the AI Gateway data retention window, since the
daily spend aggregate is not purged. `/ai/spend/users` and the export
keep the retention check, now in `aiSpendRetainedPeriod`.
- Returns `count`, `totals` (`spend_micros`, `median_spend_micros`),
`intervals` (daily spend, one interval per UTC day, zeros included), and
`users` (paginated with `limit`/`offset`, most spend first, with
per-group spend).
- Echoes the applied period, and reads every value in one read-only
repeatable-read transaction so the values agree.
- Extract `parseOrganizationAISpendPage`, shared with `/ai/spend/users`.
- Add handler tests.
Closes AIGOV-728
> [!NOTE]
> Generated by Coder Agents on behalf of @ssncferreira.
|
||
|
|
7db53f3b47 |
feat: add organization AI spend overview queries (#31711)
## Description Adds 3 database queries for the organization AI spend overview. They read the daily spend aggregate (`ai_user_daily_spend`), scoped to the organization through each spend row's effective group, over a `[period_start, period_end)` range of UTC days. 1. `GetOrganizationAISpendTotals`: the number of users with spend, the total spend, and the median of the users' total spend. 2. `ListOrganizationAISpendPerUser`: paginated users, most spend first, with one row for each group their spend was attributed to. 3. `GetOrganizationAISpendPerDay`: the spend for every UTC day in the range, including days without spend. Note: Only users with spend above zero are included, so users with only unpriced usage are not listed. ## Changes - dbauthz checks matching `/ai/spend/users` (read on organization group members). - Querier and dbauthz tests. Related to AIGOV-728 > [!NOTE] > Generated by Coder Agents on behalf of @ssncferreira. |
||
|
|
73cf6bfc69 |
feat: show the prebuilds user in the Workspaces Owner filter (#31827)
Closes CLIENT-267 All prebuilt workspaces belong to the `prebuilds` system user, and the docs tell admins to find them with `owner:prebuilds`. However, the Owner filter on the Workspaces page never offers `prebuilds`, as its options come from `GET /api/v2/users`, which hides system users. There was no way to ask for them either, as the users search rejected `include_system:true`. So, the users search now accepts `include_system:true`, and the Owner filter sends it. Without the term, `/users` behaves exactly as before, and you still only get back users you're allowed to read. The other endpoints that share the users search (org members, group members and the MCP ACL picker) now accept the term too, but they still hide system users. The template ACL picker reuses the `/users` handler, so it'll list `prebuilds` if you ask for it. <img width="697" height="313" alt="image" src="https://github.com/user-attachments/assets/afdbfe89-f74c-46c5-8675-4aab231f7e25" /> |
||
|
|
c303114fae |
perf(coderd/database): batch chat authz fetch in GetChatDiffStatusesByChatIDs (#31880)
> Generated by Coder Agents on behalf of @cstyan. |
||
|
|
871eb9bd3e |
feat(coderd/schedule/cron): add Standard schedule constructor (#32063)
## Why Automations store a five-field cron expression and an IANA time zone. The scheduler and the management API will parse both with `coderd/schedule/cron`, as the "Schedules" section of the [Automations RFC](https://app.notion.com/p/3ded579be59281e39950e686df02fa3d) says. This adds the parser they will call. Part of CODAGT-1135. ## What changed - `cron.Standard(spec, timeZone)` parses a five-field spec that runs on the wall clock of an IANA zone. Every field can be restricted. It returns an error for an empty spec, a wrong field count, a `CRON_TZ=` or `TZ=` prefix, an empty or unknown zone, a zone name with an empty, `.` or `..` segment such as `./localtime`, a name that starts with `posix/` or `right/` (directories of zone data that some hosts have, with other versions of the zones), `Local` (a Go name for the host's zone), a name whose last segment is `localtime`, `posixrules` or `Factory` (files in a host's zone data that are not zones), and a spec that never runs, such as `0 0 30 2 *`. - `Next` on a `Standard` schedule follows the RFC's daylight saving rule. A repeated local time runs once, at its first instance. A nonexistent local time runs at the first local time after the change. `robfig/cron` does neither: in America/New_York it runs `30 1 * * *` twice on 2026-11-01 and skips `30 2 * * *` on 2026-03-08. So `Standard` walks the matching wall-clock readings and resolves each one to an instant from the zone's UTC offsets. Every loop has a fixed bound, and `Next` always returns a later time for a `Standard` schedule. It searches at least 16 years ahead, past the eight years between two leap days across 2100. `Weekly`, `Daily` and `TimeRange` keep their behavior. - `Schedule.MinInterval()` returns the shortest distance between two times of day at which the schedule runs, counted around the clock. One time of day gives 24 hours. The management API will compare it with the deployment's minimum interval. ## Scope This PR adds no minimum interval policy and no configuration. Nothing calls the new code yet. The scheduler loop and the API come in later PRs. ## Tradeoffs - `MinInterval` reads only the hour and minute fields, with no calendar. The RFC allows a minimum interval of one minute to 24 hours, so the value only has to be right below 24 hours, and a scan over the calendar kept missing cases. While the zone's UTC offset does not change, no two runs are closer than the value. The doc comment states two limits, and a test pins each. - A change of the UTC offset (daylight saving) can shorten a gap that spans it by up to the size of the change: `1,58 1,3 * * *` in America/New_York gives 57 minutes, and its runs on 2026-03-08 are 3 minutes apart. - A pair of times across midnight counts even when the day fields never match two days in a row: `0,58 0,23 * * 1` gives 2 minutes, but its runs are 58 minutes apart at least, so a check can reject a schedule that the minimum allows. - The RFC rule, read as written, means a wildcard schedule such as `*/15 * * * *` does not run during the second pass of a repeated hour. In America/New_York on 2026-11-01 it runs at 01:45 EDT and next at 02:00 EST. - `Standard` resolves a reading from the zone's offsets and does not use Go's zone period bounds, which are wrong after the last transition in a zone's data. ## Live check None. This is a library that nothing calls yet. --- _Generated with [`xum`](https://github.com/coder/xum) • Model: `anthropic:claude-sonnet-5-5` • Thinking: `xhigh`_ |
||
|
|
3086d4f4c0 |
fix: run queued messages on the chat's model when theirs is unavailable (#31632)
A queued message whose model was disabled, had its provider disabled, or was deleted ran on the organization default, even when the chat was on another available model, and the chat then moved to the default. Promotion now tries the queued message's model, then the chat's last model, then the organization default, the order a message sent without a model already uses. Saving a queued edit without picking a model now sends the picker's model when the row's model is no longer selectable, as a history edit does. Refs CODAGT-971 |
||
|
|
a48a8c0cd7 |
fix(coderd): skip validation for unexpired chat git tokens (#31921)
Avoid provider validation requests when OAuthToken().Valid() is true. They count against the provider's API rate limit without changing which access token the chat resolver returns. Keep token refresh behavior for tokens within OAuth2's early-expiry margin or past expiry, and preserve provider validation for other external-auth callers. Refs CODAGT-1543 |
||
|
|
decaf6e8c5 |
fix(coderd/x/chatd): shorten tool input validation errors (#32078)
Nested builtin tool input errors repeat an absolute schema path at every recursion level. Unwrap the validator's error chain to retain only the innermost path wrapper and its cause, with a comment explaining the dependency's wrapping behavior. Also remove the redundant sentence about exact property names and duplicate keys from the retry guidance. Existing schema-validation tests now assert complete tool error messages for root failures, nested properties, arrays, and maps. Validation behavior and diagnostics embedded in the cause are preserved. Refs #31541. |
||
|
|
eb6094979b |
chore: refresh AI model price book (#32072)
## Price book changes 32 models added, 18 models removed, 42 models changed. <details> <summary>Added</summary> - anthropic/claude-fable-5-2 - azure/claude-fable-5-2 - bedrock/anthropic.claude-fable-5-2 - bedrock/global.anthropic.claude-fable-5-2 - bedrock/global.moonshotai.kimi-k3.1 - bedrock/us.anthropic.claude-fable-5-2 - bedrock/us.moonshotai.kimi-k3.1 - copilot/claude-fable-5.2 - copilot/gemini-4-carbon - google/gemini-4-carbon - openrouter/anthropic/claude-fable-5.2 - openrouter/baidu/ernie-5.5-flash-preview:free - openrouter/baidu/ernie-light-dt-1.0-intl - openrouter/google/gemini-4-carbon - openrouter/google/gemini-omni-1.1-flash - openrouter/google/gemini-omni-1.2-flash - openrouter/kt/midm-k-3.0-pro - openrouter/liquid/lfm-3-1.7b - openrouter/liquid/lfm-3-1.7b:free - openrouter/meta/muse-ember - openrouter/motif-technologies/motif-3 - openrouter/nex-agi/nex-n2.6 - openrouter/thinkingmachines/inkling-2-small:free - openrouter/unbiased/pareto-26.10-preview - vercel/alibaba/qwen3.6-35b-a3b - vercel/anthropic/claude-fable-5.2 - vercel/anthropic/claude-fable-5.2-fast - vercel/deepseek/deepseek-v4.1-flash-1017-ultrafast - vercel/google/gemini-4-carbon - vercel/google/gemini-omni-1.2-flash - vercel/humansand/ampersand-v0.1 - vercel/meta/muse-ember </details> <details> <summary>Removed</summary> - openrouter/bytedance-seed/seed-1.6 - openrouter/bytedance-seed/seed-1.6-flash - openrouter/bytedance-seed/seed-2.0-code - openrouter/inception/mercury-2 - openrouter/inception/mercury-2.5 - openrouter/inception/mercury-3 - openrouter/minimax/minimax-01 - openrouter/minimax/minimax-m1 - openrouter/minimax/minimax-m2 - openrouter/minimax/minimax-m2-her - openrouter/minimax/minimax-m2.1 - openrouter/nvidia/nemotron-3.5-content-safety:free - openrouter/qwen/qwen-2.5-7b-instruct - openrouter/stealth/crouton-alpha - openrouter/xingchen-agi/xing4.0-29b-a4b:free - vercel/bytedance/seed-1.6 - vercel/bytedance/seed-1.8 - vercel/google/gemini-3.1-flash-image-preview </details> <details> <summary>Changed</summary> - openrouter/anthracite-org/magnum-v4-72b - openrouter/deepseek/deepseek-chat - openrouter/deepseek/deepseek-v4-flash - openrouter/deepseek/deepseek-v4-flash-0731 - openrouter/deepseek/deepseek-v4-flash-vision-exp - openrouter/deepseek/deepseek-v4-pro - openrouter/deepseek/deepseek-v4-pro-0813 - openrouter/deepseek/deepseek-v4.1-flash - openrouter/google/gemma-4-26b-a4b-it - openrouter/ibm-granite/granite-4.2-8b - openrouter/inclusionai/ling-3.1-flash - openrouter/mancer/weaver - openrouter/meta-llama/llama-3.1-8b-instruct - openrouter/meta-llama/llama-3.2-3b-instruct - openrouter/meta-llama/llama-3.3-70b-instruct - openrouter/minimax/minimax-m2.5 - openrouter/minimax/minimax-m2.7 - openrouter/mistralai/mistral-small-3.2-24b-instruct - openrouter/moonshotai/kimi-k2.5 - openrouter/moonshotai/kimi-k2.7-code - openrouter/moonshotai/kimi-k3.1-pro - openrouter/nvidia/nemotron-3-super-120b-a12b - openrouter/nvidia/nemotron-3.5-super-vl - openrouter/openai/gpt-oss-120b - openrouter/qwen/qwen3-14b - openrouter/qwen/qwen3-next-80b-a3b-instruct - openrouter/qwen/qwen3.5-397b-a17b - openrouter/qwen/qwen3.5-9b - openrouter/qwen/qwen3.6-27b - openrouter/reflection/beam-1.0-501b-a23b - openrouter/undi95/remm-slerp-l2-13b - openrouter/z-ai/glm-4.5-air - openrouter/z-ai/glm-5.2 - openrouter/z-ai/glm-5.3-flash - openrouter/~anthropic/claude-fable-latest - openrouter/~deepseek/deepseek-flash-latest - openrouter/~deepseek/deepseek-pro-latest - openrouter/~deepseek/deepseek-v4-flash-latest - openrouter/~moonshotai/kimi-latest - openrouter/~z-ai/glm-flash-latest - vercel/alibaba/qwen3.8-27b - vercel/nvidia/nemotron-3.5-lightning </details> ## Review notes Regenerated by `make gen/aibridge-prices` from the live [models.dev](https://models.dev) catalog. Both artifacts come from one snapshot, so they ship together: - `coderd/aibridge/prices/data/prices.json` - `site/src/modules/aiModels/knownModels/knownModelsGenerated.json` These are customer-visible cost numbers taken from upstream data, so this PR is never merged automatically. The summary above lists what moved; check the diff for exact figures before approving. Opened automatically by the [aigateway-prices-refresh workflow](https://github.com/coder/coder/actions/runs/43188102112). Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> |
||
|
|
85e689c89c |
feat(coderd/x/chatd/chatwebhook): add bearer, github and standard_webhooks verifiers (#32062)
## Why An automation webhook must prove that a request comes from the holder of the automation's secret before Coder parses the JSON body. The approved RFC ([Automations for Coder Agents](https://app.notion.com/p/3ded579be59281e39950e686df02fa3d), Webhooks section) puts that check behind one `WebhookVerifier` interface with three schemes. This PR adds those verifiers as a pure Go package. The webhook route that calls them comes in a later PR. Part of CODAGT-1135. ## What changed New leaf package `coderd/x/chatd/chatwebhook`: - `WebhookVerifier`, with the three methods exactly as the RFC writes them, and three verifiers: `bearer` (constant-time token compare, one or more spaces after `Bearer` as in RFC 6750 section 2.1), `github` (HMAC-SHA256 of the raw body in `X-Hub-Signature-256`) and `standard_webhooks` (HMAC-SHA256 of `webhook-id.webhook-timestamp.body`, `v1,` entries in `webhook-signature`, five-minute timestamp window, `whsec_` secret). - `Scheme` with the three RFC enum strings, and `NewVerifier(Scheme)`. The package does not import `coderd/database`. - Every `Verify` failure wraps `ErrRejected`, so the route can answer `401`. Every `Verify` and `ValidateSecret` error is built from a fixed string: it never holds the secret, the received token or a signature. The package does not log. Only the `NewVerifier` error names the unknown scheme. - `GenerateSecret` reads 32 bytes from `crypto/rand` for `bearer` and `github`. For `standard_webhooks` it returns `ok=false`, and `ValidateSecret` checks the pasted `whsec_` secret. ## Scope - No database, no HTTP route, no wiring, no new Go module. Nothing imports the package yet, so `main` behaves as before with the `chat-automations` experiment off. - The map from the database enum to a verifier, and its "every enum value has a verifier" test, belong to the route PR. It can call `NewVerifier(chatwebhook.Scheme(automation.WebhookAuth))`. A `NewVerifier` error is a server error, not a `401`. - No deduplication by `webhook-id`, as in the RFC. The same valid request verifies every time. ## Tradeoffs - A pasted `whsec_` secret has no length bound. The spec describes sender secrets of 24 to 64 bytes, but it does not tell a receiver to reject others. A bound would lock out a sender that deviates, and it protects only the owner from the sender's own key. - The bearer verifier accepts one or more spaces between `Bearer` and the token, as RFC 6750 section 2.1 and the API key middleware do. It rejects a tab and an empty token. It also rejects trailing whitespace after the token when it receives it, but an HTTP request cannot deliver that: `net/http` strips trailing whitespace from the header value before the verifier runs, so such a request is accepted with the right token. - The signed Standard Webhooks content uses the `webhook-timestamp` text as received. The reference Go library signs the reformatted number instead. The two differ only for a timestamp with leading zeros, which a conforming sender does not send. - A `webhook-id` that contains `.` is rejected. The spec asks senders to avoid it, because it makes `id.timestamp.body` ambiguous. A repeated `Authorization`, `X-Hub-Signature-256`, `webhook-id` or `webhook-timestamp` line is rejected for the same reason. Several `webhook-signature` lines are read as one list. - `TestSecretComparisonsAreConstantTime` reads the package source and fails on `==`, `bytes.Equal` and similar on secrets. A correct early-exit compare gives the same answers as a constant-time one, so a request-level test cannot tell them apart. The scan is a guard against the likely mistake, not a proof. ## Verification - Test vectors come from the [GitHub docs](https://docs.github.com/en/webhooks/using-webhooks/validating-webhook-deliveries) and from `TestWebhookSign` in the [Standard Webhooks Go library](https://github.com/standard-webhooks/standard-webhooks/blob/main/libraries/go/webhook_test.go). The tests cite the URL beside each vector. - Live lane: none. No route, CLI or UI reaches this code yet, so there is no surface to exercise. --- _Generated with [`xum`](https://github.com/coder/xum) • Model: `anthropic:claude-sonnet-5-5` • Thinking: `xhigh`_ |
||
|
|
e307f1d5cf |
feat: add the user-scoped chat-automations experiment (#32064)
## Why The Automations RFC controls availability with one user-scoped experiment. Its text: "The user-scoped `chat-automations` experiment controls availability. Operators configure defaults with `--experiments` and can change per-user rules while Coder is running. Each surface checks the relevant caller or automation's owner, and experiment evaluation fails closed." The later backend PRs (management routes, the `manage_automations` tool, `Publish`) need that check first. This PR registers the experiment and proves the per-user, fail-closed check. Nothing consumes it yet. Part of CODAGT-1135 ## What changed - Registers `chat-automations` in `ExperimentsKnown` and `ExperimentsUserScoped`. An earlier experiment with this name was removed in #30412, and this PR restores none of its consumers. - Keeps it out of `ExperimentsSafe`, so `--experiments=*` does not turn it on. Automations act with their owner's authority, and a test pins the choice. - Adds the name to the `userScopedExperimentStaticCheck` ruleguard rule, so server code cannot decide it from the static list. `TestExperimentsUserScopedRuleguard` keeps the rule in step with `ExperimentsUserScoped`. - Regenerates the swagger files, `docs/reference/api/schemas.md` and `typesGenerated.ts`. - Adds tests for the real experiment name. The evaluator tests cover the startup default, stored `on`, `off` and condition rules for two users, and an off decision on a rules read error, a malformed rule, an unknown mode, an attribute read error, a condition error and a nil subject. A coderd test with a real database checks the same per-user behavior through `api.ExperimentEvaluator` and `GET /api/v2/experiments`. ## Scope No routes, tool, UI or chatd change. I added no helper: `Evaluator.Enabled(ctx, userID, ex)` already decides per user and fails closed, and API handlers reach it through `api.ExperimentEvaluator`. chatd has no evaluator handle on main, because #31327 removed the unused field. The first chatd consumer adds it together with its read. ## Blast radius With the experiment off, which is the default, behavior on main does not change. Two things become visible. `GET /api/experimental/experiments/rules` and `coder exp experiment-rules list` show a `chat-automations` row, and an administrator can store a rule for it. A deployment that already passes `--experiments=chat-automations` sees three small differences: the startup log reports an opted-in hidden experiment where it reported an unknown one, the deployment Overview page stops flagging the name as invalid, and the rules row shows the static default as on. Nothing reads the experiment yet, so no feature turns on. ## Live check A local `./scripts/develop.sh` deployment built from this head ran with no flag, with `--experiments=*` and with `--experiments=chat-automations`. `GET /api/v2/experiments` lists `chat-automations` for the owner and for a member only when the startup list or a rule turns it on, and `--experiments=*` does not turn it on. A condition rule `user.username == "member"` turned it on for the member and off for the owner without a restart, and `reset` restored the startup default. --- _Generated with [`xum`](https://github.com/coder/xum) • Model: `anthropic:claude-sonnet-5-5` • Thinking: `xhigh`_ |
||
|
|
b61ec4395e |
fix(coderd/coderdtest): read the swagger doc once in TestSwaggerEnumsNotNarrowed (#32065)
## Why
The `test-go-race-pg` job fails intermittently with `WARNING: DATA RACE`
in `swag.(*Spec).ReadDoc`, on pull requests whose changes do not cause
it. It failed CI runs 43102680650 and 43103681536 on two unrelated pull
requests. A green rerun of the same commit passes, so it hides as a
flake.
## Cause
swag's `Spec.ReadDoc` writes shared state, so parallel tests that each
call it race. #31829 addressed this by reading the spec once through
`swaggerDoc`, a `sync.OnceValue`. #30122 then added
`TestSwaggerEnumsNotNarrowed` with a direct
`apidoc.SwaggerInfo.ReadDoc()` call, which races with the tests that
still read the shared value.
## What changed
`TestSwaggerEnumsNotNarrowed` calls `swaggerDoc()`, like the other
tests. It was the only direct call outside the generated
`coderd/apidoc/docs.go`.
## Reproduction
- On `main` (
|
||
|
|
1126533c31 |
feat: export workspace last used time to Prometheus (#31346)
Add an opt-in coderd_workspace_last_used_timestamp_seconds gauge that reports workspaces.last_used_at, the value the dormancy threshold measures inactivity from. To enable it, set --prometheus-collect-workspace-stats or CODER_PROMETHEUS_COLLECT_WORKSPACE_STATS. A separate WorkspaceStats collector refreshes the gauge once a minute in the background from a new GetWorkspaceLastUsedForMetrics query. Each series has a workspace_id label, so a workspace that reuses a deleted workspace's name starts a new series. Workspaces owned by deleted users are excluded. Closes CLIENT-1065 |
||
|
|
0a7ef88af9 |
feat: add option to disable the embedded AI Gateway (#31967)
Adds `--ai-gateway-embedded-enabled` / `CODER_AI_GATEWAY_EMBEDDED_ENABLED` (default `true`) so standalone-only deployments can stop `coderd` from starting the in-memory AI Gateway. `CODER_AI_GATEWAY_ENABLED` stays the global switch: the AI Gateway entitlement, `/api/v2/ai-gateway/serve`, and cost-control metrics still depend only on it. The embedded daemon now starts only when both options are `true`. | `ENABLED` | `EMBEDDED_ENABLED` | Result | |---|---|---| | `true` | `true` | Unchanged | | `true` | `false` | No embedded gateway; standalone gateways keep working | | `false` | any | No AI Gateway | Refs AIGOV-514 ## Open questions for reviewers - **Coder Agents does not work with the embedded gateway disabled.** `chatd` still starts, but model calls fail with "AI Gateway transport factory is not configured" because there is no in-process transport. AIGOV-597 (routing Agents through standalone gateways) was canceled. The docs and option help state that Agents requires the embedded gateway. Is that acceptable, or should `chatd` also be turned off in this mode? - **Option name:** the ticket proposed `CODER_AI_GATEWAY_EMBEDDED_ENABLED` tentatively. OK to keep? ## Verification ``` $ go test ./cli/ -run 'TestServer_AIGateway' -count=1 ok github.com/coder/coder/v2/cli 0.518s $ go test ./enterprise/coderd/ -run 'TestAIGatewayServe' -count=1 ok github.com/coder/coder/v2/enterprise/coderd 0.755s ``` `TestServer_AIGatewayEmbedded` fails when the new check is removed. <details> <summary>Implementation notes</summary> - `codersdk/deployment.go`: new option and `AIBridgeConfig.EmbeddedEnabled` field; updated `--ai-gateway-enabled` description. - `cli/server.go`: outer `if Enabled` sets up cost-control metrics and the blocked-users collector (the standalone `/serve` endpoint uses these too). Inner `if EmbeddedEnabled` creates the daemon, registers the in-memory handler, and adds its cleanup. - `cli/server_test.go`: `TestServer_AIGatewayEmbedded` covers the three modes by checking for the embedded daemon's startup log. - `enterprise/coderd/aibridgeserve_test.go`: `TestAIGatewayServeEmbeddedDisabled` checks the entitlement and `/serve` still work with embedded disabled. - `docs/ai-governance/ai-gateway/standalone.md`: new "Disable the embedded gateway" section. - Remaining files are generated by `make gen`. - Found during review: keeping metrics inside the embedded check would have dropped cost-control metrics for standalone gateways. </details> --- Generated with Coder Agents; reviewed, revised, and verified by @mollymaefraser. |
||
|
|
fb12d6018f |
fix!: redact agent environment variable values in API responses (#31932)
Workspace, build, and agent API responses (`/workspaces`,
`/workspacebuilds/{id}`, `/workspaceagents/{id}`, and so on) returned
agent environment variable values in plain text. Template authors often
pass credentials such as API tokens through `coder_env`, so anyone who
could read the workspace could read those secrets.
`db2sdk.WorkspaceAgent` now replaces every non-empty value with
`***REDACTED***` and keeps the keys. This is the same marker the support
bundle (`sanitizeEnv`) and the agent `/debug/manifest` endpoint already
use. The agent still gets the real values through its manifest
(`db2sdk.WorkspaceAgentEnvironment`), and nothing in the CLI or frontend
reads these values from the API.
Every value is redacted because neither `coder_env` nor
`coder_agent.env` in the Terraform provider has a sensitive flag, and
`proto.Env` has no field to carry one. Selective redaction would need
changes to the provider, the proto, and the database.
## For changelog
- Agent environment variable values are redacted in API responses
(#31932)
> Workspace, workspace build, workspace agent, and template version API
responses now return `***REDACTED***` for every non-empty agent
`environment_variables` value. Keys and empty values are unchanged. The
agent still receives the real values, so workspaces behave as before.
>
> Migration: Update scripts and integrations that read agent environment
variable values from the API or from `coder list -o json`. Read the
values inside the workspace instead.
Closes ENG-1505 (coder/security#51)
<details>
<summary>Decision log</summary>
- Confirmed the leak on `main` (
|
||
|
|
7301cee2c4 |
fix: list every enum value the API can return in the Swagger spec (#30122)
The API reference is generated from Swagger comments and struct tags, and several `enums:"..."` tags narrowed a named enum type, so the docs omitted values the server returns. For example, `User.status` listed only `active` and `suspended`, not `dormant`, and `TemplateVersion.warnings` listed `DEPRECATED_PARAMETERS`, which is not a value of the type. Remove 34 `enums` tags that narrow or duplicate a named type so the referenced type's full value list is used. Tags that describe an intentional subset stay: ACL role fields (which omit the `""` deleted sentinel used only in update requests), `APIKey.scope`, and plain-string fields. `Template.provisioner` keeps its tag because `.swaggo` replaces `ProvisionerType` with `string`. Also fix enum defects that break code generators: - The provisioner jobs and daemons `status` query parameters listed their enum twice (Swagger 2.0 requires unique `enum` items) or with the wrong type. Both are now `[]codersdk.ProvisionerJobStatus` and `[]codersdk.ProvisionerDaemonStatus`, matching what each handler parses, including comma-separated values. The daemons route previously documented job statuses, which the handler rejects with a 500. - Replace the `_` zero placeholder of `WatchErrorCode` with `iota + 1`, so the spec no longer lists 0 or `x-enum-varnames: ["_", ...]`. Wire values are unchanged. Where the narrowed tag hid values the server returns, the SDK type now includes them: `LoginTypeOAuth2ProviderApp`, `BuildReasonAutodelete`, and `MissingTemplateParameter` widen the generated TypeScript unions. The dashboard attributes `autodelete` builds to Coder like other system builds. `APIKey.login_type` and `AuthorizationCheck.action` set explicit examples so the generated samples stay representative. `TestSwaggerEnumsNotNarrowed` fails when a property's enum list drops values of its named type (other than the 15 allowlisted intentional subsets), lists a value the type lacks, or repeats a value. Part of DOCS-1122, split from #30121. --- 🤖 Built with AI assistance. |
||
|
|
3728e7a4e1 |
perf: use the build parameter index in GetTemplateParameterInsights (#31765)
Speeds up `GetTemplateParameterInsights`, which the insights collector runs every 5 minutes on every replica. The `@>` array join could only hash on `name`, so every run scanned all of `workspace_build_parameters`. Unnesting the build IDs and joining on `(workspace_build_id, name)` lets the existing unique index serve it, and the results don't change since each build ID shows up at most once per group. On 8.4M synthetic build parameters the collector's 5 minute window went from 5,193 ms to about 31 ms. `TestGetTemplateParameterInsights` is new and passes on both the old and new SQL. This also fixes the Prometheus docs example, which summed per-replica copies of the same metric. I split the `workspace_app_stats` index into #31766 so this one has no migration and can be backported to 2.34 through 2.39 (2.34 to 2.38 without the docs hunk). The pre-push hook skipped tests on this push because of a stale `origin/main` in my checkout. An earlier full run only failed in `agent/` and `site/`. Part of [CORE-439](https://linear.app/codercom/issue/CORE-439). |
||
|
|
66dca90c91 |
fix(coderd/database/migrations): renumber the workspace_app_stats index migration to 000641 (#31997)
#31766 and #31748 both added a `000640` migration and both merged, so main currently has two migrations with the same number. Migrations fail to load with `duplicate migration file: 000640_workspace_app_stats_session_ended_at_idx.down.sql`, which breaks every test that sets up a database. This moves the `workspace_app_stats` index to `000641`. It's a rename only and `dump.sql` regenerates unchanged. `go test ./coderd/database/migrations/...` and the insights and dbrollup tests pass with it. Part of [CORE-439](https://linear.app/codercom/issue/CORE-439). |
||
|
|
78e1b280d1 |
perf(coderd/database): index workspace_app_stats by session_ended_at (#31766)
Adds an index on `workspace_app_stats(session_ended_at)`. The table had
no time index and is never purged, so the insights collector's app usage
query scanned all of it every 5 minutes on every replica. On 3M
synthetic rows that query went from 124 ms to 19 ms.
**NOTE**: the index build holds a lock that blocks app stat writes until
the migration batch commits (about 1.4 to 3.4 s per 3M rows in testing),
and active session upserts stop being HOT updates.
This is separate from #31765 so that one can be backported without a
migration. dbrollup's `UpsertTemplateUsageStats` still needs its own
query change to use this index on Postgres 17+ (noted on
[CORE-776](https://linear.app/codercom/issue/CORE-776)). The migration
number may need a bump at merge time.
I skipped the pre-push hook for this push since `make pre-push` also
fails on clean `main` in my workspace. This branch only failed in tests
it doesn't touch.
<details><summary>Failing pre-push tests</summary>
Clean `main` (
|
||
|
|
af7cd184c8 |
chore: point charm.land/fantasy at coder/fantasy#85 (#31746)
Part of CODAGT-1430 Relates to CODAGT-1541 Stacked on #31743. This bumps fantasy to pick up coder/fantasy#84 and coder/fantasy#85. A Chat Completions tool message can only hold text, so fantasy sends tool result media (images, audio, and now PDFs from #31384) in a `role: "user"` message right after it. That gives whatever the tool returned the authority of the user's own words, which Codex flagged on #31383 as a prompt injection risk for MCP tools that return untrusted files. coder/fantasy#84 starts that user message with a text part naming the tool call and saying the content is tool output, not instructions from the user. It only changes the Chat Completions paths (`openai` without the Responses API, `openaicompat`, OpenRouter and Vercel); Anthropic and Responses keep the media inside the tool result. AI Bridge records the last user message's text as the prompt, so on the step after a tool returns media, it'll now record the label. I think that's fine for now. coder/fantasy#85 fixes the output token limit field. The OpenAI provider sent `max_tokens` unless the model ID looked like a reasoning model, and reasoning models reject it: ``` Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead. ``` That breaks any reasoning model whose ID fantasy doesn't recognise, e.g. an Azure deployment name, which doesn't say what model it serves (Azure providers start using this client in #31952). The OpenAI provider now sends `max_completion_tokens` on every Chat Completions request, which is what OpenAI deprecated `max_tokens` in favour of. `openaicompat`, OpenRouter and Vercel keep `max_tokens`, so self-hosted servers behind the OpenAI Compatible provider are unaffected. I've added a test that checks the field on each route, and the models docs no longer say Max Completion Tokens is only for reasoning models. |
||
|
|
d4edf1631a |
fix: replay OpenAI web searches inline without item IDs (#31743)
Closes CODAGT-1425 Stacked on #31386. coder/fantasy#78, which is already on `main`, fixed the searches with no reasoning before them by sending them inline without their `id`. Stored chats still send a search as an `item_reference` when a reasoning item came right before it, though. The API pairs any stored ID with that reasoning item and rejects the whole request if it's missing, and a reference doesn't save any tokens over an inline item anyway. This bumps fantasy to pick up coder/fantasy#77, which just never references a search, so stored chats now replay earlier searches inline, like unstored ones. It also drops the `id` from replayed computer calls, which have the same problem. I've had to stack this on the CODAGT-1430 PRs, as coder/fantasy#77 merged after coder/fantasy#73 and coder/fantasy#74, so any pin that has it has them too. `TestOpenAIResponsesNoStaleWebSearchReplay` now covers a stored chat, and I deleted `TestOpenAIResponsesFullReplayPairsReasoningAndWebSearch`, as it asserted the order of references we no longer send. |
||
|
|
661b7db7b5 |
feat(coderd/x/chatd/mcpclient): count only text toward the in-process tool result cap (#31386)
Part of [CODAGT-1430](https://linear.app/codercom/issue/CODAGT-1430). Stacked on #31385. Inline and in-process MCP servers cap each tool result at 256 KiB, and media counts at its base64 size, so a PDF over about 190 KiB becomes an error. In-process servers are our own code running in coderd, so we'll count only their text toward the cap, and leave their media to the 10 MiB tool media cap in chatloop. Inline servers are untrusted, so they still count base64 media. |
||
|
|
e0e6accdd9 |
feat: return raw MCP results from workspace MCP tools (#31385)
Part of [CODAGT-1430](https://linear.app/codercom/issue/CODAGT-1430). Stacked on #31384. Workspace MCP tool results go through the agent's `convertResult`, which turns an embedded resource into a `[embedded resource: ...]` string and drops structured content, so a PDF from a workspace MCP server never reaches the model. Remote MCP tools already go through `mcpclient`'s converter, which keeps both. When a call-tool request sets `include_result`, the agent now returns the MCP `CallToolResult` as-is in `Result` and leaves `Content` empty. chattool always sets it and converts the result with `mcpclient.ConvertCallResult`, so workspace and remote MCP tools share one converter. Older agents ignore the flag and send the legacy `Content`, which we adapt to MCP content and pass to the same converter. The workspace MCP docs get a short Tool results section too. |
||
|
|
7a12022f3d |
fix: use one route-aware media rule for files and tool results (#31384)
Part of [CODAGT-1430](https://linear.app/codercom/issue/CODAGT-1430). Stacked on #31383. We had two rules for which media a model can take. `AcceptsFilePartMediaType` filtered user attachments, whilst `ToolResultMediaOmission` filtered tool results with its own lists, and only let images through on Anthropic. Now that fantasy converts tool result media like user files, the two should agree, so we'll keep one rule. `chatprovider.Model` now carries the configured provider type, and `AcceptsFilePartMediaType` reads a table keyed on the route, i.e. the configured provider plus the transport. `replaceUnsupportedMedia` applies it to user file parts and tool result media alike, including the pending tail of the turn prompt. As a result, tool result PDFs now reach Anthropic, and the 5 MiB inline image cap only applies to images. Azure OpenAI Chat Completions rejects file content parts with a 400 (`This model does not support file content types`, see https://github.com/Azure/azure-sdk-for-net/issues/51873), and Gemini's OpenAI-compatible endpoint ignores them. chatd reaches both through the openai-compat client, so the rule rejects PDFs there, and the model gets a text note instead of a failed request or a silently dropped file. Unsupported tool result media now gets the same note fantasy writes ("The tool returned `<type>` content, which cannot be displayed."), so the model sees one wording whichever layer drops it. I've also added a Tool results section to the MCP servers docs, and fixed the File attachments docs, which said every rejected attachment becomes text. Only text files do; a rejected PDF becomes a note that names it. I've also tested this end to end on Bedrock. I ran chatd from this branch with an Anthropic provider pointed at our AI gateway's Bedrock route (InvokeModel, us-east-2), plus an MCP tool that returns a PDF with a code word inside. Claude Haiku 5.5 and Claude Opus 5.5 (adaptive thinking) both read the code word from the tool result, and read it again on the next turn, when the PDF is replayed from history. The request carries the PDF as a `document` block inside `tool_result`, the same body chatd sends to Anthropic. |
||
|
|
5cbf086517 |
feat: send tool result media to providers like user files (#31383)
Part of [CODAGT-1430](https://linear.app/codercom/issue/CODAGT-1430). This is the first of four stacked PRs. Today a PDF in an MCP tool result reaches no provider. On Anthropic we strip it in chatd. On OpenAI Chat Completions fantasy swaps it for a "cannot be displayed" note, and on Responses it sends a placeholder that points at a user message it never sends. coder/fantasy#73 and coder/fantasy#74 make fantasy convert tool result media with the converter it already uses for user files: an Anthropic document inside `tool_result`, an `input_file` inside `function_call_output` on Responses, and a user file part straight after the tool message on Chat Completions. Both have merged, and this PR pins `charm.land/fantasy` to the coder_2_33 tip that has them. Chat Completions file parts now refuse text files, as Responses already did, so `AcceptsFilePartMediaType` follows. The converters take a filename, so an MCP embedded resource blob is now named after the last element of its URI path, and we store the name with the tool result so later turns send the same one. OpenAI picks the file type from the extension, so we append the right one when it's missing or wrong. We drop names with control characters, as Postgres `jsonb` rejects NUL, and we redact the name like the rest of the tool response, since a URI like `https://host/sse/<token>` would otherwise put the token in the chat history. There's also a test proving aibridge's Chat Completions re-encode keeps a user file part, as that's where these PDFs end up. Anthropic still strips tool result PDFs until the next PR in the stack. |
||
|
|
1ec94c74dc |
fix(coderd): return 500, not 403, when AI Gateway authorization hits a database error (#31617)
When coderd's `IsAuthorized` hits a database error, the AI Gateway answers 403. A coderd outage therefore looks to clients like a permission problem. The gateway can't tell coderd's own failures from rejected requests, because coderd returns both as plain errors. Worse, the API key and user lookups turn any database error into `ErrUnknownKey` or `ErrUnknownUser`. coderd now marks each database failure in `IsAuthorized` with the dRPC error code `proto.ErrCodeInternal`, where the failure happens. This covers the key, user, workspace, organization membership and effective group lookups. The key and user lookups return `ErrUnknownKey` and `ErrUnknownUser` only for `sql.ErrNoRows`. Rejections carry no code. The gateway answers a coded error with 500 and every other error with 403. dRPC sends the code with the error, so the protocol version doesn't change. A gateway and coderd on different versions keep today's 403, because the older side neither sets nor reads the code. A connection that drops during the `IsAuthorized` call still gets a 403. That's left for a follow-up. `TestAuthorization_InternalErrors` covers each database failure and each rejection on the direct, workspace token and delegated paths. `TestServeHTTP_FailureModes` covers the gateway's status codes. Refs AIGOV-737 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> |
||
|
|
5563b7444d |
fix: keep in-flight builds from blocking the prebuild hard limit (#30060)
Fixes #30058. Fixes an issue where in-flight prebuilds prevent presets from reaching their failure hard limit. `GetPresetsAtFailureLimit` now considers only completed builds (`succeeded`, `failed`, `canceled`) when evaluating consecutive failures. Pending and running builds no longer mask earlier failures, while successful and canceled builds still break the failure streak. ### Tests - Added query tests covering in-flight builds with and without a reached failure limit. - Added a reconciler test confirming hard-limited presets do not create additional prebuilds. - Fixed the test helper to properly mark successful jobs as completed. - Verified the new tests fail before the fix and pass afterward. - All relevant tests and `golangci-lint` pass. **Live validation:** A preset with builds failing after 90s (longer than the 60s reconcile interval) stopped after 4 failures, with no additional builds during a 20-minute observation. Previously, the same scenario produced 156 failed builds in approximately 2 hours. ### AI disclosure Written with AI assistance (Claude Code). I reviewed the changes, ran the tests, and verified the fix on a live deployment. Co-authored-by: Bryson Henneberger <591079+PushTheLimit@users.noreply.github.com> Co-authored-by: Bobby Ho <bobby@coder.com> |