Wires chat lifecycle hooks into chatd, gated by the
`agent-lifecycle-hooks` experiment. Part of the lifecycle hooks stack
(#27401, #27428, #27430). See `docs/admin/setup/chat-lifecycle-hooks.md`
for the consumer-facing contract.
## Summary
When a hook URL is configured, chatd dispatches `session_start`,
`user_prompt_submit`, `pre_tool_use`, `post_tool_use`, `pre_compact`,
`post_compact`, and `stop` events to the consumer and applies its
responses.
## Design
- **Stateless**: Coder stores no hook dispatch or decision state.
Delivery is at least once; consumers deduplicate on stable payload
identifiers (chat ID, event type, tool-use ID) and answer duplicates
with the same decision.
- **Admission-time prompt effects**: `user_prompt_submit` dispatches
exactly once per submission (create, send, queue, edit, subagent spawn)
and folds its effects into the stored prompt as typed message parts:
original-or-overridden user parts, then model-only `hook-context`, then
a user-visible `hook-notice`. Hook context is stripped from every
client-facing conversion; hook notices are excluded from model prompts.
The server rejects hook parts in client-submitted content.
- **Tool gating**: `pre_tool_use` allow can override tool input; deny
becomes a synthetic denied tool result, with any returned model context
persisted as a model-only transcript row so it never reaches clients.
The denial text identifies an external policy (the deployment's
lifecycle hook) as the source and marks the decision as persistent, so
the model explains the denial instead of retrying it or misreporting it
as an infrastructure failure.
- **Fail closed**: a dispatch failure rejects the triggering request or
moves the chat to the error state in the same transaction as the
affected step, so a runnable state is never published with unapproved
content.
- **Admission before persistence**: `pre_tool_use` is dispatched for the
calls the model produced, before the assistant message is stored. See
"Staged tool admission" below.
- **Fresh dispatch per tool call**: every non-provider-executed tool
call is decided by its own `pre_tool_use` dispatch; Coder never reuses
an earlier decision on the consumer's behalf. Retries re-dispatch the
same logical event.
## Structure
All hook dispatch flows through one seam: entry points build a
`chathooks.Chat` (chat identity) and a `chathooks.Message` (event
details) and call `Trigger.Trigger`, the only component that talks to
the dispatcher. The integration lives in the `coderd/x/chatd/chathooks`
subpackage, split by responsibility:
- `trigger.go`: the trigger seam; builds the wire envelope per event,
normalizes deny into a typed error, and holds the package's single
enabled-check.
- `effects.go`: pure conversion of hook results into transcript rows and
prompt parts.
- `errors.go`: failure classification (dispatch error messages, denial
mapping, tool-result dispatch-failure scanning).
- `tooluse.go`: the tool-call gate (`pre_tool_use` preflight,
`post_tool_use` payloads, applying admitted input to the step).
Server-bound glue stays in `coderd/x/chatd/hook_server.go`: the
chat-parking dispatch error handlers, the step-commit row insertion
wrappers, and the dynamic post-tool-use state loader, which depends on
chatd validation types.
This PR adopts the `codersdk/x/agenthooks` and
`coderd/x/agenthooks/dispatch` import paths introduced at the tip of
#27401; intermediate commits still reference the pre-move paths and are
not individually buildable.
## Staged tool admission
`pre_tool_use` originally ran at tool execution time, which is after the
assistant message carrying the tool call was already committed. An
`input_override` therefore had to rewrite stored message content in
place. @hugodutka pointed out that chatd treats message content as
immutable, and that the rewrite was a shortcut rather than a
requirement.
It was also a correctness problem in its own right: the rewrite only
updated the database, so the transcript could show one input while a
different one had executed.
The hook now runs before the step is persisted:
```text
provider stream ends (tool calls complete, in memory)
-> pre_tool_use dispatch per call
-> ONE transaction: assistant row with admitted inputs, synthetic denials, hook rows
-> execute
```
The step is inserted once, carrying the input the tool runs with.
`UpdateChatMessageContentByID` and `Tx.UpdateMessageContent` are deleted
from #27428, so message content stays immutable.
Two consequences, both intentional:
- **Clients converge rather than wait.** Tool-call parts still stream
live, so a rewritten call briefly shows the model's proposed input
before the committed message replaces it. The chat store already clears
stream state when an assistant message arrives, so the stored input wins
with no frontend change and no added latency before tool cards appear.
- **A call already in history was already admitted.** Execution consumes
the stored input instead of dispatching a second decision, which keeps
one dispatch and one set of hook effects per call. A consumer policy
change between admission and execution applies to later calls, not to
calls already admitted.
The per-chat debug endpoint still records the provider's original tool
input. Its purpose is to report provider behavior, and it requires an
explicit per-chat debug flag; the invariant here covers the transcript.
## Configuration
Adds `chat-hook-url`, `chat-hook-secret`, `chat-hook-timeout`, and
`chat-hook-enabled` deployment options with startup validation. The
flags are hidden from `coder server --help` while the feature is
experimental; the setup guide documents them.
## Tool input validation
Built-in tool arguments reach a consumer as raw JSON with key spelling
preserved, but the tools decode those bytes with Go, which matches
struct fields case-insensitively and keeps the last match. A policy
reading `path` could therefore authorize one value while the tool
executed another, and a lone case variant such as `{"PATH":"/secret"}`
was invisible to a policy checking for `path`.
Coder now rejects a built-in tool call whose input repeats a key or
spells a schema property with different capitalization, before the
`pre_tool_use` dispatch, so a consumer is never asked to authorize bytes
whose meaning depends on the reader. Rejected calls produce an error
result the model can retry; unambiguous calls in the same batch still
run. A consumer-authored `input_override` is rechecked after the
dispatch and fails the turn closed, because the model cannot correct it.
Dynamic and MCP inputs are excluded because the client and the workspace
agent execute those calls rather than coderd.
Two paths needed more than a schema check. Execution resolves a
deprecated tool name to its canonical tool, so validation resolves
aliases first. The `edit_files` decoder also reads `search` and
`replace`, which its schema does not advertise, so those aliases are now
matched exactly and their case variants ignored.
A hook denial now returns a structured 403 carrying `kind:
"hook_denied"`, mirroring the dispatch-failure response that already
carries its own kind. Without it a client cannot tell a policy decision
apart from a generic failure, and the chat UI titled a denial "Request
failed". Adding a kind needs no migration: `ChatErrorKind` is persisted
only inside the JSONB `chats.last_error` column, whose decoder accepts
unknown kinds.
The hook docs also correct the tool-input convergence window. A batch
dispatches sequentially before the assistant row commits, so the
original input stays visible for a span that scales with the number of
tool calls in the step rather than a single hook timeout.
> This PR was written by Mux, an AI coding agent, on Mike's behalf.
Normalizes non-standard code-fence language tags across `docs/**` so a
strict highlighter (Shiki, used by Fumadocs) won't fail the build on an
unrecognized language, and unifies redundant synonym tags onto one
canonical form per language. The current renderer (Speed-Highlight)
detects the language from the code content, not the fence label, so this
drift wasn't visible until now.
## Changes
- `hcl` -> `tf` (199 fences, including indented ones nested in
numbered/bulleted lists). Shiki ships `hcl` and `terraform` as two
distinct grammars (not aliases); every `hcl`-tagged fence in `docs/**`
is actually Terraform resource/data/provider syntax, so the more
specific `terraform` grammar is correct for all of them. `tf` is Shiki's
own alias for that grammar, and it's also what GitHub's own markdown
renderer resolves to the same HCL/Terraform highlighting.
- `pwsh`/`powershell` -> `ps1`. Both `ps` and `ps1` are registered
PowerShell aliases in Shiki, but on GitHub's renderer only `.ps1` is a
registered file extension (`.ps` isn't), so `ps1` renders identically to
`powershell` there today while bare `ps` would silently lose
highlighting.
- `env` -> `dotenv` (a dedicated Shiki grammar for `KEY=VALUE` files)
- `text`/`output`/`none`/`url` -> `txt`. Same built-in plain-text
fallback either way, just shorter.
- `Dockerfile` -> `dockerfile` (lowercase)
- `bash`/`shell` -> `sh` (732 fences). Shiki and GitHub both alias all
three to a single shell grammar; this was already the style guide's
stated preference, just not enforced across the existing corpus until
now.
- `markdown` -> `md` (4 fences). Alias of the same grammar in both Shiki
and GitHub.
- `jsonc` -> `json` (1 fence). The block has no comments or trailing
commas, so it doesn't need the comments-capable grammar.
- `ts` -> `tsx` (2 fences, `docs/about/contributing/frontend.md`).
Verified the actual content tokenizes identically under both grammars,
and a sibling block in the same file already needs `tsx` for real JSX,
so unifying to one tag is safe for this file. Documented a caveat: `tsx`
mis-tokenizes the legacy angle-bracket type-assertion syntax
(`<Type>value`), which is invalid in real `.tsx` files anyway, so use
`value as Type` instead.
- `yml` -> `yaml` (1 fence)
- Updated `docs/.style/style-guide/formatting.md` to document all
canonical tags
`promql` (2 fences) and `caddyfile` (2 fences) are left as-is. Shiki
doesn't bundle a grammar for either, so they need a custom grammar
registration when the site adopts Shiki, rather than degrading to `txt`.
Tracked as follow-up work under DOCS-118 and
[DOCS-544](https://linear.app/codercom/issue/DOCS-544/vendor-a-local-promql-grammar-for-shiki-syntax-highlighting)
(promql).
Does not touch `offlinedocs/`.
Linear:
[DOCS-476](https://linear.app/codercom/issue/DOCS-476/normalize-docs-code-fence-languages-de-risk-shikifumadocs)
<details>
<summary>How the fence tags were verified</summary>
Each tag was tested against a real `shiki@latest` highlighter instance
(`codeToHtml`/`codeToTokens`) and cross-checked against GitHub's
`@wooorm/starry-night` grammar sources (the renderer that actually
displays these `.md` files today, in repo browsing and PR diffs), since
that's what determines whether brevity is safe before Shiki adoption:
```text
FAIL env -- Language `env` is not included in this bundle.
FAIL Dockerfile -- Language `Dockerfile` is not included in this bundle.
FAIL promql -- Language `promql` is not included in this bundle.
FAIL caddyfile -- Language `caddyfile` is not included in this bundle.
FAIL pwsh -- Language `pwsh` is not included in this bundle.
FAIL output -- Language `output` is not included in this bundle.
```
`hcl` doesn't error in Shiki, since it's a real grammar, but that's
exactly the trap: it was silently rendering every fence with the generic
HCL grammar instead of the Terraform-specific one. Every `hcl`-tagged
fence in `docs/**` was manually checked against `origin/main` and is
genuinely Terraform content.
For `ts`/`tsx`, tokenizing the actual doc content confirmed identical
output under both grammars; a synthetic test with the legacy
angle-bracket cast syntax confirmed `tsx` degrades on that specific
construct, which the style guide now calls out.
The first normalization pass only matched fence tags at column 0
(`^```tag$`), missing tags indented inside numbered/bulleted lists. A
follow-up pass caught the remaining occurrences at any indentation
level.
</details>
---
*This PR description and the underlying changes were prepared with Coder
Agents assistance.*
The `/api/v2/csp/reports` endpoint is unauthenticated and CSRF-exempt,
since it's the browser's `report-uri` target, and decoded request bodies
with no size limit. This let an attacker post arbitrarily large JSON
bodies to force unbounded heap allocation and OOM the server (Cure53
CDM-02-007).
Wraps the request body in `http.MaxBytesReader` before decoding and
returns 413 when the limit is exceeded, matching the existing convention
used by `files.go`, `aitasks.go`, and `exp_chats.go`.
Fixes: https://github.com/coder/security-disclosures/issues/171
Adds `--aigateway-proxy-target` option to
`deploymentGroupAIGatewayProxy` that defines URL to which intercepted
requests should be forwarded to.
Forward URL used to be hardcoded to `coderAPI.AccessURL` pointing to
embedded Gateway. With addition of standalone AI Gateway this needs to
be configurable.
Renamed `aibridgeproxyd.Server.coderAccessURL` and `coderAccessPort` ->
`gatewayURL` and `gatewayPort` + option to better reflect reality.
This PR adds a new opt-in setting, `CODER_OIDC_REDIRECT_ALLOWED_HOSTS`,
that lets a single Coder deployment complete OIDC login on more than one
hostname. When the allowlist is non-empty, Coder picks the OIDC
`redirect_uri` based on the incoming request's Host header (validated
against the list) instead of always using the static URL derived from
`CODER_ACCESS_URL`. When unset, the (default) behavior is identical to
today.
The motivation is that a single Coder deployment is frequently reachable
via multiple hostnames - for example, an internal hostname for users on
a corporate VPN and a different hostname routed through a zero-trust
gateway for users off-VPN - but OIDC login today only works on whichever
single hostname `CODER_ACCESS_URL` points to, because the `redirect_uri`
sent to the IdP is fixed at server startup. Users who reach the
deployment on any other valid hostname can see the login page but fail
the OIDC callback, since the IdP redirects them back to a hostname they
can't reach (or whose cookies they don't have).
<!-- Authored by Coder Agents on behalf of @Emyrk. -->
Adds an opt-in `CODER_DANGEROUS_OIDC_EMAIL_FALLBACK` flag (alias
`--dangerous-oidc-email-fallback`) for IdP brokers that do not issue a
stable `sub` for the same user across connections.
relates to GRU-69
Modifies replicasync to handle discovering NATS enabled primary replicas explicitly, and passing that info to the NATS Pubsub.
This PR adds a new deployment value to explicitly represent the host or IP that the replica can be reached on. It isn't wired up to the CLI, but piggybacks on the DERP config for now.
We learn the NATS port directly from NATS at runtime, and propagate it thru replicasync to learn all peers for clustering.
Add a periodic purge job for `boundary_logs` rows past their retention
threshold, following the same pattern as the existing audit log and
connection log purge jobs in `dbpurge`.
Expose a `--boundary-log-retention` deployment flag (env
`CODER_BOUNDARY_LOG_RETENTION`, YAML `retention.boundary_logs`). Default
is `0` (keep indefinitely). When set to a positive duration, `purgeTick`
deletes rows where `captured_at` is older than the threshold in batches
of 10,000, matching other log purge operations. The `boundary_logs`
label is added to the `records_purged_total` Prometheus counter.
Also removes the random-UUID fallback for `OwnerID` in
`dbgen.BoundarySession`. The previous fallback generated a UUID that
could never satisfy the `boundary_sessions_owner_id_fkey` FK constraint,
masking test setup bugs. Callers must now provide a valid user ID or
accept NULL (the legitimate "user deleted" state).
Replace the env-based `BuildProviders` with a DB-backed loader. The database is now the single source of truth for runtime provider configuration; env config arrives via `SeedAIProvidersFromEnv` (run at boot) and `BuildProviders` reads it back as `aibridge.Provider` instances. `cli/server.go` and `enterprise/cli/server.go` both call the same path, so aibridged and aibridgeproxyd see the same provider set.
Per-provider `DumpDir` is replaced by a top-level `CODER_AI_GATEWAY_DUMP_DIR` base; each provider's effective dump path is `<base>/<provider name>`.
> AI tools where used when creating this PR
This PR removes environment variable parsing from `/aibridge` directory.
Added env variables/flags for dump dir as coder options.
Only added to new indexed provider options
(`CODER_AIBRIDGE_PROVIDER_<N>_*`) not to deprecated legacy env variables
(`CODER_AIBRIDGE_ANTHROPIC_*` and `CODER_AIBRIDGE_OPENAI_KEY_*`).
Reverted adding `MaxRetries` option as it will be removed soon due to
key failover work:
https://github.com/coder/coder/pull/24783#discussion_r3155544808
## Summary
Adds `--ai-gateway-allow-byok` deployment option to control whether
users can use Bring Your Own Key (BYOK) mode with AI Gateway.
When disabled (`--ai-gateway-allow-byok=false`), BYOK requests are
rejected with a 403 and a message directing the admin to enable the
flag. Centralized key authentication works regardless of this setting.
Defaults to `true` (BYOK allowed).
---------
Co-authored-by: Danny Kopping <danny@coder.com>
_Disclaimer: produced mostly by Claude Opus 4.6 following detailed
planning._
## Summary
- Support multiple instances of the same AI Bridge provider type via
indexed env vars (`CODER_AIBRIDGE_PROVIDER_<N>_<KEY>`), following the
`CODER_EXTERNAL_AUTH_<N>_<KEY>` pattern
- Existing single-provider env vars (`CODER_AIBRIDGE_OPENAI_KEY`, etc.)
continue to work unchanged
- Setting both a legacy env var and an indexed provider with the same
name errors at startup to prevent silent misconfiguration
- Mark legacy provider fields (`OpenAI`, `Anthropic`, `Bedrock`) as
deprecated in `AIBridgeConfig` in favor of `Providers`
## Example
```sh
CODER_AIBRIDGE_PROVIDER_0_TYPE=anthropic
CODER_AIBRIDGE_PROVIDER_0_NAME=anthropic-corp
CODER_AIBRIDGE_PROVIDER_0_KEY=sk-ant-corp-xxx
CODER_AIBRIDGE_PROVIDER_0_BASE_URL=https://llm-proxy.internal.example.com/anthropic
CODER_AIBRIDGE_PROVIDER_1_TYPE=anthropic
CODER_AIBRIDGE_PROVIDER_1_NAME=anthropic-direct
CODER_AIBRIDGE_PROVIDER_1_KEY=sk-ant-direct-yyy
```
Each instance is routed by name:
- /api/v2/aibridge/**anthropic-corp**/v1/messages
- /api/v2/aibridge/**anthropic-direct**/v1/messages
Closes
[AIGOV-157](https://linear.app/codercom/issue/AIGOV-157/spike-to-understand-if-there-is-a-simple-way-to-handle-multi-api-key)
---------
Signed-off-by: Danny Kopping <danny@coder.com>
## Description
Blocks `CONNECT` tunnels to private and reserved IP ranges in
aibridgeproxyd, preventing the proxy from being used to reach internal
networks.
The Coder access URL is always exempt (hostname+port match) so the proxy
can reach its own deployment. It is possible to exempt additional ranges
via `CODER_AIBRIDGE_PROXY_ALLOWED_PRIVATE_CIDRS`.
DNS rebinding is handled differently per path:
* Direct (no upstream proxy): validate the resolved IP right before the
TCP dial, no window between check and connect.
* Upstream proxy: Resolves and checks before forwarding to the upstream
dialer. A small rebinding window exists since the upstream proxy
re-resolves independently.
## Changes
* Add blocked IP denylist covering private, reserved, and
special-purpose ranges
* Add `AllowedPrivateCIDRs` option with CLI flag and env var
* Wire IP checks into `proxy.ConnectDial` for both upstream and direct
paths
* Add tests for blocked/allowed cases across direct dial, upstream
proxy, CIDR exemptions, and CoderAccessURL exemption
Notes: documentation will be handled in a follow-up PR.
Closes: https://github.com/coder/security/issues/124
## Summary
- add a hidden deployment config option for chat acquire batch size
(`CODER_CHAT_ACQUIRE_BATCH_SIZE` / `chat.acquireBatchSize`)
- thread the configured value into chatd startup while preserving the
existing default of `10`
- clamp the deployment value to the `int32` range before passing it into
chatd
- regenerate the API/docs/types/testdata artifacts for the new config
field
## Why
`chatd` currently acquires pending chats in batches of `10` via a
compile-time default. This change makes that batch size
operator-configurable from deployment config, so we can tune acquisition
behavior without another code change.
- Adds `_API_BASE_URL` to `CODER_EXTERNAL_AUTH_CONFIG_`
- Extracts and refactors existing GitHub PR sync logic to new packages
`coderd/gitsync` and `coderd/externalauth/gitprovider`
- Associated wiring and tests
Created using Opus 4.6
## Description
Adds optional TLS support for the AI Bridge Proxy listener. When TLS cert and key files are provided, the proxy serves over HTTPS instead of plain HTTP.
## Changes
* New configuration options to enable TLS on the proxy listener
* Wraps the TCP listener in `tls.NewListener` when configured
* Tests for validation errors, invalid files, and full integration (tunneled + MITM) through a TLS listener
Note: Documentation for TLS listener setup and client configuration will be handled in a follow-up PR.
Related to: https://github.com/coder/internal/issues/1335
If a deployment has 2 domains, overriding the oidc url allows the oidc
redirect to differ from the access_url
response to https://github.com/coder/coder/discussions/21500
**This config setting is hidden by default**
## Summary
Add circuit breaker support for AI Bridge to protect against cascading
failures from upstream AI provider rate limits (HTTP 429, 503, and
Anthropic's 529 overloaded responses).
## Changes
- Add 5 new CLI options for circuit breaker configuration:
- `--aibridge-circuit-breaker-enabled` (default: false)
- `--aibridge-circuit-breaker-failure-threshold` (default: 5)
- `--aibridge-circuit-breaker-interval` (default: 10s)
- `--aibridge-circuit-breaker-timeout` (default: 30s)
- `--aibridge-circuit-breaker-max-requests` (default: 3)
- Update aibridge dependency to include circuit breaker support
- Add tests for pool creation with circuit breaker providers
## Notes
- Circuit breaker is **disabled by default** for backward compatibility
- When enabled, applies to both OpenAI and Anthropic providers
- Uses sony/gobreaker internally via the aibridge library
## Testing
```
make test RUN=TestPoolWithCircuitBreakerProviders
```
## Description
Adds upstream proxy support for AI Bridge Proxy passthrough requests.
This allows aiproxy to forward non-allowlisted requests through an
upstream proxy. Currently, the only supported configuration is when
aiproxy is the first proxy in the chain (client → aiproxy → upstream
proxy).
## Changes
* Add `--aibridge-proxy-upstream` option to configure an upstream
HTTP/HTTPS proxy URL for passthrough requests
* Add `--aibridge-proxy-upstream-ca` option to trust custom CA
certificates for HTTPS upstream proxies
* Passthrough requests (non-allowlisted domains) are forwarded through
the upstream proxy
* MITM'd requests (allowlisted domains) continue to go directly to
aibridge, not through the upstream proxy
* Add tests for upstream proxy configuration and request routing
Closes: https://github.com/coder/internal/issues/1204
## Description
Implements selective MITM (Man-in-the-Middle) in `aibridgeproxyd` so
that only requests to allowlisted domains are intercepted and decrypted.
Requests to all other domains are tunneled directly without decryption.
## Changes
* New config option: `CODER_AIBRIDGE_PROXY_DOMAIN_ALLOWLIST` (default:
`api.anthropic.com`,`api.openai.com`)
* Selective MITM: Uses `goproxy.ReqHostIs()` to only intercept `CONNECT`
requests to allowlisted hosts
* Certificate caching: Now only generates/caches certificates for
allowlisted domains
* Validation: Startup fails if domain allowlist is empty or contains
invalid entries
Closes: https://github.com/coder/internal/issues/1182
Closes https://github.com/coder/coder/issues/21360
A few considerations/notes:
- I've kept the number of conns to 10 in all other places, except coderd
- which uses the config value
- I opted to also make idle conns configurable; the greater the delta
between max open and max idle, the more connection churn
- Postgres maintains a [_process_ per
connection](https://www.postgresql.org/docs/current/connect-estab.html),
contrary to what the comment said previously
- Operators should be able to tune this, since process churn can
negatively affect OS scheduling
- I've set the value to `"auto"` by default so it's not another knob one
_has to_ twiddle, and sets max idle = max conns / 3
---------
Signed-off-by: Danny Kopping <danny@coder.com>
Because this affects more than just the template insights
page (specifically it also affects the deployment stats endpoint which
is shown on bottom bar and Prometheus), the group is being renamed
generically to just "stats collection". In the future if we need to
affect the other stats we can put those options here.
Then, because this change only affects a portion of stats, specifically
usage stats like connection and application time, bytes sent, etc, add a
new sub-group called "usage stats".
Then finally add back the "enable" flag. This also gives us a place to
one day place an "anonymize" flag if we need to go that route.
## Description
Adds the core AI Bridge MITM proxy daemon. This proxy intercepts HTTPS traffic, decrypts it using a configured CA certificate, and forwards requests to AIBridge for processing.
## Changes
* Added `aibridgeproxyd` package with the core proxy server implementation
* Added configuration options: `CODER_AIBRIDGE_PROXY_ENABLED`, `CODER_AIBRIDGE_PROXY_LISTEN_ADDR`, `CODER_AIBRIDGE_PROXY_CERT_FILE`, `CODER_AIBRIDGE_PROXY_KEY_FILE`
* Added tests for server initialization and MITM functionality
Closes https://github.com/coder/internal/issues/1180
**Breaking Change:** Existing oauth apps might now use PKCE. If an
unknown IdP type was being used, and it does not support PKCE, it will
break.
To fix, set the PKCE methods on the external auth to `none`
```
export CODER_EXTERNAL_AUTH_1_PKCE_METHODS=none
```
Closes#20399
To summarize the original commit messages:
- Do not log stats to the database.
- Return errors on the insight endpoints.
- Update the frontend to show those errors.
- Also fixes an issue with getting the user status count via codersdk,
since I added a test to ensure it was not disabled by this flag and it
was sending the wrong payload.
## Summary
This adds configurable overload protection to the AI Bridge daemon to
prevent the server from being overwhelmed during periods of high load.
Partially addresses coder/internal#1153 (rate limits and concurrency
control; circuit breakers are deferred to a follow-up).
## New Configuration Options
| Option | Environment Variable | Description | Default |
|--------|---------------------|-------------|---------|
| `--aibridge-max-concurrency` | `CODER_AIBRIDGE_MAX_CONCURRENCY` |
Maximum number of concurrent AI Bridge requests. Set to 0 to disable
(unlimited). | `0` |
| `--aibridge-rate-limit` | `CODER_AIBRIDGE_RATE_LIMIT` | Maximum number
of AI Bridge requests per second. Set to 0 to disable rate limiting. |
`0` |
## Behavior
When limits are exceeded:
- **Concurrency limit**: Returns HTTP `503 Service Unavailable` with
message "AI Bridge is currently at capacity. Please try again later."
- **Rate limit**: Returns HTTP `429 Too Many Requests` with
`Retry-After` header.
Both protections are optional and disabled by default (0 values).
## Implementation
The overload protection is implemented as reusable middleware in
`coderd/httpmw/ratelimit.go`:
1. **`RateLimitByAuthToken`**: Per-user rate limiting that uses
`APITokenFromRequest` to extract the authentication token, with fallback
to `X-Api-Key` header for AI provider compatibility (e.g., Anthropic).
Falls back to IP-based rate limiting if no token is present. Includes
`Retry-After` header for backpressure signaling.
2. **`ConcurrencyLimit`**: Uses an atomic counter to track in-flight
requests and reject when at capacity.
The middleware is applied in `enterprise/coderd/aibridge.go` via
`r.Group` in the following order:
1. Concurrency check (faster rejection for load shedding)
2. Rate limit check
**Note**: Rate limiting currently applies to all AI Bridge requests,
including pass-through requests. Ideally only actual interceptions
should count, but this would require changes in the aibridge library.
## Testing
Added comprehensive tests for:
- Rate limiting by auth token (Bearer token, X-Api-Key, no token
fallback to IP)
- Different tokens not rate limited against each other
- Disabled when limit is zero
- Retry-After header is set on 429 responses
- Concurrency limiting (allows within limit, rejects over limit,
disabled when zero)
Adds `--disable-workspace-sharing` option.
Workspace sharing is disabled by not including user and group ACLs in
the workspace RBAC object, which prevents ACL-based authz.
Closes https://github.com/coder/internal/issues/1072
The commit also adds saving of workspace user/group ACLs in the test DB
data generator.
Replace hardcoded 7-day retention for workspace agent logs with
configurable retention from deployment settings. Defaults to 7d to
preserve existing behavior.
Depends on #21038
Updates #20743
Add `RetentionConfig` with server flags for configuring data retention:
- `--audit-logs-retention`: retention for audit log entries
- `--connection-logs-retention`: retention for connection logs
- `--api-keys-retention`: retention for expired API keys (default 7d)
Updates #20743
Currently, when AI Bridge is enabled AND the `oauth2` and
`mcp-server-http` experiments are enabled we inject Coder's MCP tools
into all intercepted AI Bridge requests.
This PR introduces a config to control this behaviour.
**NOTE:** this is a backwards-incompatible change; previously these
tools would be injected automatically, now this setting will need to be
explicitly enabled.
---------
Signed-off-by: Danny Kopping <danny@coder.com>
The authz recorder is causing a lot of memory to be allocated, and is a
memory leak for websocket connections.
This change makes it opt-in on a per request basis (ontop of `isDev`).
To get the authz headers, use `Copy as cURL` on chrome and append the
header `x-authz-checks=true`.
Solves #15575
Adds OAuth access token revocation when unlinking external auth
provider. Due to revocation not being consistently implemented by
providers this is only best effort attempt. Unsuccessful revocation
won't influence link removal.
Michael Suchacz