mirror of
https://github.com/coder/coder.git
synced 2026-09-22 05:05:20 +08:00
Michael Suchacz
c17bed25e0
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.
About
Coder is a self-hosted platform for running AI coding agents and cloud development environments on infrastructure you control. It works with any cloud, IDE, OS, Git provider, and IDP.
Coder Workspaces
Coder Workspaces are cloud development environments defined with Terraform, connected through a secure Wireguard tunnel, and automatically shut down when not in use. Agents and developers share the same workspace infrastructure.
- Defined in Terraform: Templates describe the infrastructure for each workspace, from EC2 VMs and Kubernetes Pods to Docker containers.
- Any architecture and OS: Support ARM and x86-64 across Windows, Linux, and macOS from a single deployment.
- Managed by admins: Platform teams create and maintain templates that enforce approved images, resource limits, and security policies.
- Accessed from any IDE: Connect through VS Code, JetBrains, Cursor, a web terminal, remote desktop, or SSH.
- Automatic shutdown: Idle workspaces stop automatically to reduce cloud spend, and restart in seconds when needed.
Coder Agents
Coder Agents is a native AI coding agent built into Coder. The agent loop runs in the Coder control plane on your infrastructure, not in the workspace and not in a vendor's cloud. Developers interact with agents through the web UI or the REST API for programmatic and CI-driven workflows.
- Self-hosted agent loop: The control plane handles planning, model calls, and tool dispatch. Workspaces have zero AI awareness.
- No API keys in workspaces: LLM credentials stay in the control plane.
- Any model: Anthropic, OpenAI, Google, Bedrock, or self-hosted endpoints. Switching is a configuration change.
- Governance and cost controls: Centralized model approval, per-user spend limits, and audit logging.
- Open source and inspectable: The full platform is available to audit and extend.
IDE support
You can use:
-
Any Web IDE, such as
- code-server
- JetBrains Projector
- Jupyter
- And others
-
Your existing remote development environment:
-
A file sync such as Mutagen
Why remote development
Provisioning consistent development environments for a large engineering team is difficult. Each developer has preferences for operating systems, editors, and toolchains, and ensuring a reliable build environment across all of them is a maintenance burden. A missed step during onboarding or an unsupported local configuration can cost hours of debugging.
Remote development solves this by moving the environment off the developer's machine and into managed infrastructure. The developer's laptop becomes a portal into the actual compute where work happens. If a device is lost or replaced, access is simply revoked; no source code or credentials are stored locally.
This approach provides:
- Speed: Server-grade hardware accelerates builds, tests, and large workloads without requiring expensive local machines.
- Consistency: Infrastructure tools such as Terraform, nix, Docker, and Dev Containers produce identical environments for every developer.
- Security: Source code stays on private servers. Users and groups are managed through SSO and RBAC.
- Compatibility: Workspaces share infrastructure configurations with staging and production, reducing configuration drift.
- Accessibility: Browser-based IDEs and remote IDE extensions let developers work from any device, including lightweight laptops, Chromebooks, and tablets.
Read more on the Coder blog, the Slack engineering blog, or from Alex Ellis at OpenFaaS.
Why Coder
The key difference between Coder and other platforms is that the entire system, agent loop, control plane, model routing, and workspace provisioning, runs on infrastructure you control.
For agents, this means platform teams can:
- Run the entire agent loop on their infrastructure, with no SaaS dependency for orchestration.
- Define MCP servers, skills, and system prompts centrally so every agent session starts with the same tools, policies, and context.
- Keep LLM credentials out of workspaces entirely.
- Tie every agent action to an authenticated user identity.
- Support air-gapped and restricted-network deployments with self-hosted models.
For workspaces, this means admins can:
- Support any architecture (ARM, x86-64) and operating system (Windows, Linux, macOS).
- Modify pod/container specs, such as adding disks, managing network policies, or setting/updating environment variables.
- Use VM or dedicated workspaces, developing with Kernel features (no container knowledge required).
- Enable persistent workspaces, which are like local machines, but faster and hosted by a cloud service.
Pricing
Coder is free and open source under the GNU Affero General Public License v3.0. All developer productivity features are included in the open source version. A Premium license is available for enhanced support and custom deployments.
How Coder works
Coder workspaces are represented with Terraform, but you do not need to know Terraform to get started. The Coder Registry provides production-ready templates for AWS EC2, Azure, Google Cloud, Kubernetes, and other providers.
Providers and compute environments
Workspaces can include more than just compute. Terraform can add storage buckets, secrets, sidecars, and other resources.
See the templates documentation for details.
What Coder is not
-
Coder is not an infrastructure as code (IaC) platform.
- Terraform is the first IaC provisioner in Coder, allowing Coder admins to define Terraform resources as Coder workspaces.
-
Coder is not a DevOps/CI platform.
- Coder workspaces can be configured to follow best practices for cloud-service-based workloads, but Coder is not responsible for how you define or deploy the software you write.
-
Coder is not an online IDE.
- Coder supports common editors, such as VS Code, vim, and JetBrains, all over HTTPS or SSH.
-
Coder is not a collaboration platform.
- You can use Git with your favorite Git platform and dedicated IDE extensions for pull requests, code reviews, and pair programming.
-
Coder is not a SaaS/fully-managed offering.
- Coder is a self-hosted solution. You must host Coder in a private data center or on a cloud service, such as AWS, Azure, or GCP.