Adds a new documentation page at `docs/ai-coder/agents.md` describing
Coder Agents — the built-in chat interface, API, and lightweight AI
coding agent that runs in the Coder control plane.
## What's included
- Overview of what Coder Agents is and who it's for (regulated
industries, platform teams, existing Coder deployments)
- How the architecture works (agent loop in coderd, outbound to LLM
providers, connects to workspaces via existing daemon connection)
- Key features: automatic template/workspace selection, sub-agents, chat
persistence, message queuing
- Security benefits of the control plane architecture (no API keys in
workspaces, simpler network boundaries, centralized enforced control,
user identity attached)
- LLM provider support table (verified against
`coderd/chatd/chatprovider/chatprovider.go`)
- Built-in tools reference
- Comparison to Coder Tasks
- Product status (internal preview, early access next)
## Summary
The macOS `.dylib` is only used by Coder Desktop macOS v0.7.2 or older.
v0.7.2 was released in August 2025. v0.8.0 of Coder Desktop macOS, also
released in August 2025, uses a signed Coder slim binary from the
deployment instead.
It's unlikely customers will be using Coder Desktop macOS v0.7.2 and the
next release of Coder simultaneously, so I think we can safely remove
this process, given it slows down CI & release processes.
## Changes
- **Makefile**: Remove `DYLIB_ARCHES`, `CODER_DYLIBS` variables and
`build/coder-dylib` target
- **scripts/build_go.sh**: Remove `--dylib` flag and all dylib-specific
logic (c-shared buildmode, CGO, plist embedding, vpn/dylib entrypoint)
- **scripts/sign_darwin.sh**: Remove dylib-specific comment
- **CI (ci.yaml)**: Remove `build-dylib` job, artifact download/insert
steps, and `build-dylib` dependency from `build` job
- **Release (release.yaml)**: Remove `build-dylib` job, artifact
download/insert steps, and `build-dylib` dependency from `release` job
- **vpn/dylib/**: Delete entire directory (`lib.go` + `info.plist.tmpl`)
- **vpn/router.go, vpn/dns.go**: Clean up comments referencing dylib
The slim and fat binary builds are completely unaffected — the dylib was
an independent build target with its own CI job.
_Generated by mux but reviewed by a human_
## Problem
On the `/agents/:agentId` detail page, text typed into the chat input
was lost when navigating away and returning. The empty-state page
(`/agents`) already persisted drafts via `localStorage`, but individual
conversation pages did not.
## Solution
Adds per-conversation draft persistence to `useConversationEditingState`
in `AgentDetail.tsx`, following the same patterns used elsewhere in the
agents page:
- Drafts are stored under `agents.draft-input.<chatID>` keys
- The saved draft is read as the editor's initial value on mount
- `localStorage` is updated on every content change
- The key is removed when the input is cleared or a message is sent
successfully
Fixescoder/internal#642
We recently fixed Windows specific flakes for this test and reenabled
it. It then failed intermittently due to context deadline expiration.
The temporary path created on Windows contained invalid characters. This
resulted in a silent startup script failure on Windows. The test then
fruitlessly waited until context expiration. The test now uses a valid
path on Windows.
## Problem
Two bugs in the agents chat flow:
1. **Optimistic rendering glitch**: When sending a message while the
agent is busy, a fake message with a negative ID appears in the
timeline, then gets rolled back to the queued state. This causes a
jarring flash.
2. **Auto-promoted messages not appearing**: When the server
auto-promotes a queued message after finishing a task, the promoted user
message doesn't show up in the timeline until the LLM finishes its
response.
## Root Causes
**Bug 1**: The optimistic rendering system injected placeholder messages
with `id: -Date.now()` into the store. When the server responded with
`queued: true`, the optimistic message was rolled back — but the user
had already seen it flash in the timeline.
**Bug 2**: In `processChat`'s deferred cleanup, the auto-promoted
message was published via `publishEvent()`, which only delivers to local
in-process stream subscribers. The SSE subscriber goroutine only
forwards `message_part` events from the local channel — it ignores
`message` events. Durable events reach the SSE client via pubsub → DB
read, but `publishEvent` doesn't trigger a pubsub notification. The
explicit `PromoteQueued` endpoint correctly used `publishMessage()`
(which does both), but the auto-promote path did not.
## Changes
### Frontend (`site/`)
- **AgentDetail.tsx**: Remove optimistic message injection from send and
edit flows. Instead, use the `CreateChatMessageResponse.message` from
the POST response to insert the real server message into the store
immediately.
- **ChatContext.ts**: Remove the negative-ID cleanup logic from
`upsertDurableMessage` that stripped optimistic placeholders when real
messages arrived.
- **chatStore.test.ts**: Remove 2 tests for negative-ID optimistic
message behavior.
### Backend (`coderd/chatd/`)
- **chatd.go**: In `processChat` cleanup, replace `publishEvent()` with
`publishMessage()` for auto-promoted messages. This ensures the pubsub
notification (`AfterMessageID`) is sent, so SSE subscribers read the new
message from the DB immediately.
I'm having a hard time reproducing [this
Heisenbug](https://github.com/coder/internal/issues/1154) in PR CI, but
it seems to happen pretty often on main, so I would like to add some
logging for a few more page events to the ones we already have.
closes https://github.com/coder/internal/issues/464
# Summary
This PR resolves a flaky test that was sensitive to DST transitions in
various time zones. The root of the flake was:
* a bug; the query and its tests assume 24 hours per day
* the tests used local system time, which resulted in failures for dates
proximal to DST transitions
# Changes
Query:
The original query assumed 24 hour intervals between each day, which is
not a valid assumption. It now increments `1 day` at a time.
Database tests:
Database level tests for the query all assumed 24 hour days. They now
increment in DST-aware days instead. Instead of using time.Now() as a
base for testing, the test uses a series of dates over the course of an
entire year, to ensure that DST transition dates are present in every
test run.
# API Endpoint
The endpoint that delivers the user status chart now accepts an IANA
timezone name as a parameter and passes it, keeping the existing offset
as a fallback, to the database query.
API level tests were added to ensure the correct response form and error
behaviour. Correctness of content is tested at the database level.
## Summary
Replace hand-coded per-provider field components, form state types,
validation schemas, and builder functions with generic schema-driven
code that reads from the auto-generated
`chatModelOptionsGenerated.json`.
## Changes
### `ModelConfigFields.tsx` (492 → 341 lines)
- Remove 6 per-provider components (`OpenAIFields`, `AnthropicFields`,
`GoogleFields`, `OpenAICompatFields`, `OpenRouterFields`,
`VercelFields`)
- Remove exported option arrays (`modelConfigReasoningEffortOptions`,
etc.)
- Add `renderSchemaField()` that dispatches to
`InputField`/`SelectField`/`JSONField` based on `field.input_type` from
the generated schema
- `ModelConfigFields` now calls `getVisibleProviderFields()` instead of
a switch statement
- `GeneralModelConfigFields` now calls `getVisibleGeneralFields()`
instead of hard-coding 6 InputField instances
### `modelConfigFormLogic.ts` (742 → 525 lines)
- Remove 6 per-provider form state types and empty defaults
- Remove 6 per-provider Yup validation schemas
- Remove 6 per-provider builder functions (`buildOpenAIOptions`, etc.)
- Remove 2 switch-case dispatch blocks (validation + build)
- Add `buildEmptyProviderState()` that walks schema fields to create
empty form state
- Add schema-driven `extractModelConfigFormState()` and
`buildModelConfigFromForm()`
- Add `yupTestForField()` + `buildYupSchema()` generating Yup validation
from field metadata
- Lazy-cache per-provider Yup schemas for performance
### `modelConfigFormLogic.test.ts`
- All 83 tests updated for the new nested state shape
- Uses `toContain` for error message assertions since labels now come
from schema descriptions
## Motivation
The auto-generated schema (`chatModelOptionsGenerated.json`) was merged
in #22568 but not yet consumed by the UI. This PR wires it up so that
when a new provider or field is added in Go (`codersdk/chats.go`),
running `make gen` regenerates the JSON schema and the UI automatically
picks up the new fields — no manual TypeScript changes needed.
**Production code reduced from 1234 to 866 lines (-30%).**
## Problem
The subscribe flow in `useWebpushNotifications` called
`pushManager.subscribe()` without first requesting the `Notification`
permission. When the browser permission state is `"denied"` (e.g. from a
previous prompt dismissal), the browser throws:
```
DOMException: Registration failed - permission denied
```
This surfaced as a confusing error toast on the agents page. The error
has nothing to do with Coder RBAC roles — it's the browser denying the
push subscription because notification permission was previously
declined. An admin who had granted browser permission wouldn't see this;
a user who previously dismissed or denied the prompt would.
## Fix
Added an explicit `Notification.requestPermission()` call before
`pushManager.subscribe()`. This:
1. **Re-prompts** the user if the permission state is `"default"` (not
yet decided)
2. **Throws a clear, actionable error** if the permission is `"denied"`:
*"Notifications are blocked by your browser. Please allow notifications
for this site in your browser settings."*
3. **Only proceeds** to `pushManager.subscribe()` after permission is
confirmed as `"granted"`
## Tests
New test file `useWebpushNotifications.jest.ts`:
- **requests notification permission before subscribing** — verifies
`requestPermission()` is called before `pushManager.subscribe()`
- **throws a clear error when permission is denied** — verifies the
user-friendly error message
- **does not call pushManager.subscribe when permission is denied** —
verifies we bail out early
On the `/agents` page, the "View Workspace" link in the header dropdown
menu was navigating in the same tab via `navigate()`. This changes it to
`window.open(workspaceRoute, "_blank")` so it opens in a new browser
window/tab instead.
It's frustrating when I want to view my workspace and then I have to go
back and find my chat.
## Problem
There is a race condition in the chat stream reconnect path. When a
client connects (or reconnects) to `/stream`, sometimes they only see a
`status: running` event but never receive any `message_part` events —
the stream appears stuck.
## Root Cause
In `processChat`, the sequence is:
1. `publishStatus(running)` — broadcasts `status: running` to all
subscribers and via pubsub.
2. `runChat()` is called.
3. Inside `runChat`, there's significant setup work (model resolution,
DB queries, title generation, prompt building, instruction resolution).
4. Only **after** all that setup does `runChat` set `buffering = true`
on the stream state.
If a client connects to `/stream` between steps 1 and 4:
- `Subscribe()` reads `chat.Status == running` from the DB, so it
includes `status: running` in the snapshot.
- But `buffering` is still `false`, so `subscribeToStream` returns an
**empty** local snapshot (no message_parts).
- `publishToStream` **drops** all `message_part` events when `buffering`
is false.
- Result: client sees `running` but never gets any streaming content.
## Fix
Move the `buffering = true` setup (and its deferred cleanup) from
`runChat` into `processChat`, right before `publishStatus(running)`.
This guarantees the buffer is active before any subscriber can observe
`status: running`, so:
- The snapshot always includes any in-flight `message_part` events.
- `publishToStream` never drops parts because buffering is already on.
Despite the SDK type having an `Archived` field for chats, this data was
never fetched from the database — the `GetChatsByOwnerID` query
hardcoded `AND archived = false`, and the `convertChat` function never
mapped the field.
This PR adds an optional `archived` query parameter to `GET
/api/experimental/chats`:
| Value | Behavior |
|-------|----------|
| *(not provided)* | Returns all chats (active and archived) |
| `archived=false` | Returns only non-archived chats |
| `archived=true` | Returns only archived chats |
This follows the same pattern used by template versions
(`sqlc.narg('archived')` nullable boolean).
Also fixes `convertChat` to populate the `Archived` field in API
responses, which was never being set despite existing on the SDK type.
The search input was removed from `AgentsSidebar` but the
`ArchivedAgentsSearchAutoExpands` story still referenced the `Search
agents...` placeholder, causing the Storybook interaction test to fail:
```
within(<div#storybook-root>).getByPlaceholderText("Search agents...")
Unable to find an element with the placeholder text of: Search agents...
```
This PR removes the stale story.
This is an attempt to address coder/internal#1154
Tests appear to fail often on `verifyParameters`, which asserts input
visibility and value in series for all expected parameters. This change
makes the same assertions in parallel, hopefully completing before
timeout.
Add Prometheus metrics to the boundary log proxy for observability:
- batches_dropped_total (reason: buffer_full, forward_failed)
- logs_dropped_total (reason: buffer_full, forward_failed,
boundary_channel_full, boundary_batch_full)
- batches_forwarded_total
Also add BoundaryStatus to the BoundaryMessage envelope so boundary
can report dropped log counts as a separate wire message. The agent
records these as Prometheus metrics, making boundary-side data loss
visible. Backwards compatibility for older versions of boundary is maintained.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Adds database columns and server-side logic to track interception lineage via tool call IDs. When an interception ends, the server resolves the correlating tool call ID to find the parent interception and links them via `parent_id`.
New `provider_tool_call_id` column on `aibridge_tool_usages` and `parent_id` column on `aibridge_interceptions`, with indexes for lookup. `findParentInterceptionID` queries by tool call ID and filters out the current interception to find the parent.
Adapted from the [coder/coder `dk/prompt_provenance_poc`](https://github.com/coder/coder/compare/main...dk/prompt_provenance_poc) branch.
Depends on [coder/aibridge#188](https://github.com/coder/aibridge/pull/188).
Closes https://github.com/coder/internal/issues/1334
Adds database columns and server-side logic to track interception lineage via tool call IDs. When an interception ends, the server resolves the correlating tool call ID to find the parent interception and links them via `parent_id`.
New `provider_tool_call_id` column on `aibridge_tool_usages` and `parent_id` column on `aibridge_interceptions`, with indexes for lookup. `findParentInterceptionID` queries by tool call ID and filters out the current interception to find the parent.
Adapted from the [coder/coder `dk/prompt_provenance_poc`](https://github.com/coder/coder/compare/main...dk/prompt_provenance_poc) branch.
Depends on [coder/aibridge#188](https://github.com/coder/aibridge/pull/188).
Closes https://github.com/coder/internal/issues/1334
Removes the search input and restyles the New Agent button in the agents
sidebar:
- Removed the search input box
- Replaced the outlined button with a subtle, left-aligned button
featuring a `SquarePenIcon`
- Button icon and text alignment matches the tree node items in the
sidebar
<img width="769" height="337" alt="image"
src="https://github.com/user-attachments/assets/2284c8c0-6294-4823-9ce0-5cc72b0d0054"
/>
relates to #21335
Enables the agent socket by default and updates docs to strike references to having to enable it.
The PRs in this stack change the MCP server that Tasks use to update their status to rely on the agent socket, rather than directly dialing Coderd with the agent token.
Default disable was a reasonable default when it was only used for the experimental script ordering features, but now that we want to use it for Tasks, it should be default on.
## Problem
The **Open in Cursor** and **Open in VS Code** buttons on the agent
detail page were broken. Clicking them did nothing.
### Root Cause
The `handleOpenInEditor` handler in `AgentDetail.tsx` called
`window.location.assign()` with a custom protocol URI (`vscode://` or
`cursor://`) **after** an `await API.getApiKey()` call. This creates an
async boundary that breaks the browser's user gesture chain, causing
custom protocol navigations (`vscode://`, `cursor://`) to be silently
blocked by the browser.
The handler was invoked from a Radix `DropdownMenuItem.onSelect`, which
adds another layer of event indirection that makes the gesture chain
more fragile.
In contrast, the workspace page's `VSCodeDesktopButton` works because it
uses a direct `onClick` handler on a button element.
## Fix
- **Eagerly fetch and cache the API key** via `useQuery` when workspace
and agent data is available
- **Make `handleOpenInEditor` synchronous** — it reads the cached key
instead of awaiting a network call, keeping `window.location.assign()`
within the original user gesture context
- **Disable buttons** while the API key is still loading
(`canOpenEditors` now gates on key availability)
- **Simplify** the `onOpenInEditor` callback (remove `void` async
wrapper)
Prebuilds need to be valid. Before this change, you can push a template
version that's preset will fail when making a prebuild. This PR ensures
all presets that are used for prebuilds are valid
## Problem
Flaky test:
`TestCloseDuringShutdownContextCanceledShouldRetryOnNewReplica`
(coder/internal#1371)
The test intermittently fails because the chat ends up in `waiting`
status instead of `pending` after server shutdown.
## Root Cause
There is a race condition in `processChat` where `runChat` completes
successfully just as the server context is being canceled during
`Close()`. The sequence:
1. Server calls `Close()`, canceling the server context.
2. The LLM HTTP response has already been fully written by the mock
server (the stream closes normally before context cancellation
propagates to the HTTP client).
3. `runChat` returns `nil` (success) instead of `context.Canceled`.
4. The existing `isShutdownCancellation` check only runs when `runChat`
returns an error, so the shutdown is not detected.
5. `processChat`'s deferred cleanup marks the chat as `waiting` instead
of `pending`.
6. The test's assertion that the chat is `pending` never becomes true.
This race is timing-dependent — it only triggers when the mock server's
HTTP response completes in the narrow window between context
cancellation being initiated and it propagating through the HTTP
transport layer.
## Fix
Add a server context check after `runChat` returns successfully. If the
server is shutting down (`ctx.Err() != nil`), override the status to
`pending` so another replica can pick up the chat.
This is the same pattern already used for the error path
(`isShutdownCancellation`), extended to cover the success path.
Extend the wire protocol for the boundary <-> agent unix socket with
a message envelope.
The envelope creates a boundary <-> agent data path that is separate
from the agent <-> coderd path. This lets boundary send operational
metadata (drop counts, configuration like jail type, capabilities)
that the agent can act on locally (e.g. Prometheus metrics) or use
to enrich outbound requests, without polluting the coderd-facing proto
with fields coderd never consumes.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
## Summary
Removes `time.Sleep` calls in two test files by replacing them with
deterministic or event-driven alternatives.
### Changes
**`coderd/provisionerjobs_test.go`** (34.5s → 0.25s)
Replaced `time.Sleep(1500ms)` with a direct SQL `UPDATE` to bump
`created_at` by 2 seconds. The sleep existed purely to ensure different
timestamps for sort-order testing. The fix is deterministic and cannot
flake. Uses `NewDBWithSQLDB` (the test already required real Postgres
via `WithDumpOnFailure`).
**`coderd/database/pubsub/pubsub_test.go`** (2.05s → 1.3s)
Replaced `time.Sleep(1s)` with a `testutil.Eventually` retry loop that
publishes and checks for subscriber receipt. This is the idiomatic
pattern in the codebase. The old sleep waited for pq.Listener to
re-issue LISTEN after reconnect; the new code polls until it actually
works.
## Summary
Change the four main `coderdtest` Await helper functions to poll at
`IntervalFast` (25ms) instead of `IntervalMedium` (250ms):
- `AwaitTemplateVersionJobCompleted`
- `AwaitWorkspaceBuildJobCompleted`
- `WorkspaceAgentWaiter.WaitFor`
- `WorkspaceAgentWaiter.Wait`
These are called **~855 times** across the test suite. Each call
previously wasted ~125ms on average waiting for the next poll tick.
`AwaitTemplateVersionJobRunning` already used `IntervalFast` — this
makes all Await helpers consistent.
## Measured Impact
Local benchmarks (postgres, `-short -count=1 -p 8 -parallel 8
-tags=testsmallbatch`):
| Package | Before | After | Delta |
|---|---|---|---|
| enterprise/coderd | 90.8s | 76.0s | **-16.3%** |
| coderd | 65.6s | 56.5s | **-13.8%** |
| cli | 57.9s | 37.8s | **-34.7%** |
| enterprise (root) | 41.1s | 39.9s | -2.9% |
| **Sum of all packages** | **623s** | **543s** | **-12.8%** |
Zero test failures across all 199 packages.
The pause/resume endpoints were only registered under /api/experimental
but the frontend and Go SDK were calling /api/v2, resulting in 404s.
Register the routes in the v2 group, update the SDK client paths, and
fix swagger annotations (Accept → Produce) since these POST endpoints
have no request body.
This pull-request cleans up various issues we I noticed while using the
`<Checkbox />` element.
* Margins were missing around the outsides of the `<Checkbox />`
components.
* Resolved the story of the `WithLabel` so things are lined up.
* Lowered the `borderRadius` down to `2px` (This can be cleaned up by
Tailwind 4 later).
* Refined checkbox styling across focus, disabled, checked, and
indeterminate states to be inline with Figma.
* Simplified checkbox indicator rendering and centered the icon with
absolute positioning.
- Unexposes port 5432 so it doesn't conflict with existing databases.
You can still hop into the DB if you need.
- Updates multiple CODER_ envs to sensible defaults, overrideable via env
This pull-request follows up #22060
Felt wrong to only make use of Geist when there is a Monospace variant
here too. Felt best we default to this as the default font as its inline
with the rest of the application. This also updates the lower line for
Workspace Statistics 🙂
This pull-request updates our icons to be inline with the Figma file.
They were slightly too small in the two variants of `--avatar-default`
and `--avatar-sm`. Now these are inline with what we have defined and
using the correct variants in the breadcrumbs.
Matt Vollmer