mirror of
https://github.com/coder/coder.git
synced 2026-11-02 05:23:10 +08:00
Michael Suchacz
8eb1e43f0b
## Problem
Selecting a reasoning effort on a Google (Gemini) chat model either got
dropped or broke the request outright, and even when the request
succeeded, Gemini chats never showed thinking blocks:
- The native `fantasygoogle` path had no case in `applyReasoningEffort`
at all, so effort was silently ignored. With only `include_thoughts`
configured, Gemini 3+ models default to no thinking, so chats produced
zero reasoning tokens.
- The path chat generation actually uses today routes Google providers
through aibridge as openai-compat clients and forwarded the raw effort
string as `reasoning_effort`. Google's OpenAI-compatible endpoint
validates that value against the model's supported thinking levels
instead of clamping, so out-of-range efforts fail the whole request with
HTTP 400. Verified against the live Gemini API: `gemini-3.1-pro-preview`
accepts low/medium/high and rejects none/minimal/xhigh/max. Because the
config's default effort is sent whenever the user selects nothing, a
config whose default falls outside the model's supported set failed
every no-selection message.
- Google's OpenAI-compatible endpoint never emits thought text for
`reasoning_effort` requests. Thoughts require
`extra_body.google.thinking_config` with `include_thoughts`, which is
mutually exclusive with `reasoning_effort`, and even then the thought
deltas come back inline in `content` (marked
`extra_content.google.thought`, wrapped in `<thought>` markers) rather
than in the `reasoning_content` field OpenAI-compatible clients parse.
So no reasoning parts were ever produced, persisted, or rendered for
Gemini chats (verified live in UAT).
## Fix
Reasoning effort:
- Map the global reasoning effort scale onto Google's `thinking_level`
in `applyReasoningEffort`, gated to Gemini 3+ models and clamped to each
model's supported subset (3.0 Pro launched with LOW/HIGH only, 3.1 Pro
added MEDIUM, Flash accepts all four through 3.6 while 3.7 and the
latest flash aliases drop MINIMAL, image models accept HIGH plus MINIMAL
for flash). Pre-Gemini-3 models reject `thinking_level`, so effort
degrades to a no-op for them and a config-pinned level is dropped at
call time instead of failing every generation.
- Add `thinking_level` to `ChatModelGoogleThinkingConfig` so admins can
pin a level for configs without user-selectable effort. A resolved
per-turn effort overrides the pinned level; an explicit
`thinking_budget` wins over both since the Google API rejects requests
that set `thinking_budget` and `thinking_level` together, and config
validation now rejects that combination up front. The admin panel picks
up the new field automatically through the generated model-options
schema.
Thinking blocks on the openai-compat path (the path chat generation
uses), translated in both directions at the existing Gemini transport
seam:
- chatd's OpenAI-compat request patch swaps `reasoning_effort` for an
equivalent `extra_body.google.thinking_config` with `include_thoughts`
enabled: `thinking_level` on Gemini 3+ (clamped to the model's supported
subset), `thinking_budget` on earlier models (effort `none` maps to
budget 0, which disables thinking). Non-Gemini models and non-Google
routes are untouched.
- The same transport rewrites Google's responses so thought output
surfaces as `reasoning_content`, which fantasy already converts into
reasoning parts: streaming bodies are rewritten per SSE line (preserving
streaming latency), JSON bodies in place, with the
`<thought>`/`</thought>` markers stripped at the transitions.
- aibridge's chat-completions interception preserves the `extra_body`
passthrough that openai-go's typed params silently drop, forwarding it
to Google upstreams only, so the chatd -> aibridge -> Google route
carries the thinking config end to end.
## Validation
- `go test ./coderd/x/chatd/... ./aibridge/... ./internal/googleopenai
./codersdk` pass; golangci-lint clean on touched packages.
- New unit tests cover the effort-to-level mapping for every effort
value, per-model clamping on both paths, the effort-to-thinking_config
request swap (level, budget, pinned-config precedence, non-Gemini
passthrough), SSE and JSON thought-response rewriting (fixtures mirror
live captures, including tool-call and no-thought streams), an
end-to-end fake-SSE test through the fantasy pipeline proving reasoning
deltas emerge with markers stripped, and aibridge `extra_body`
preservation (including proof the typed round trip drops it).
- Verified live against `generativelanguage.googleapis.com` on
`gemini-3-flash-preview` and `gemini-2.5-flash` through the real
chatprovider stack: reasoning deltas stream through with markers
stripped and answer text intact.
- Live UAT on a dogfood deployment (remote Coder Agents chat, 4 rounds):
effort clamping verified end to end (rounds 1-2), the missing-thoughts
limitation isolated (round 3), and the fix validated (round 4):
expandable, readable thinking blocks render for
`gemini-3.1-pro-preview`, `gemini-3-flash-preview`, and
`gemini-2.5-flash` at effort=high; no `<thought>` marker leaks;
effort=none produces no block and no 400; multi-turn history replay
works with a thinking block on each turn.
- `make gen` regenerated `typesGenerated.ts` and
`chatModelOptionsGenerated.json`; `pnpm -C site lint:types` passes.
> 🤖 Mux acted on Mike's behalf to create this pull request.
<!-- mux-attribution: model=claude-opus-4-6 thinking=high -->
Coder is a self-hosted platform for cloud development environments and AI coding agents. Workspaces are defined with Terraform, connected through a secure Wireguard® tunnel, and automatically shut down when not used. Coder Agents runs a native AI coding agent whose loop executes in the control plane on your infrastructure, with no API keys in workspaces.
- Define cloud development environments in Terraform
- EC2 VMs, Kubernetes Pods, Docker Containers, etc.
- Automatically shutdown idle resources to save on costs
- Onboard developers in seconds instead of days
- Delegate coding work to AI agents on your infrastructure
- Bring any model (Anthropic, OpenAI, Google, Bedrock, self-hosted)
- No LLM credentials in workspaces, user identity on every action
- Centralized model governance, cost tracking, and audit logging
Quickstart
Check out our install guides for other methods and a complete tutorial.
Try Coder with the install script on Linux and macOS, or grab the latest binary or installer from GitHub Releases on Windows:
curl -L https://coder.com/install.sh | sh
Start the server and open http://localhost:3000 to create your initial user, create a Docker template, and provision your first workspace:
coder server
For a production deployment, add a PostgreSQL database (version 13 or later) and an external access URL, and see our validated architectures for sizing and infrastructure guidance:
coder server --postgres-url <url> --access-url <url>
Without these flags, Coder uses a built-in database and sets up a *.try.coder.app access URL for evaluation. Use coder --help for the full list of flags and environment variables.
Documentation
Browse the documentation or visit a specific section below:
- Workspaces: Workspaces contain the IDEs, dependencies, and configuration information needed for software development
- Templates: Templates are written in Terraform and describe the infrastructure for workspaces
- Coder Agents: Delegate coding work to AI agents running on your self-hosted infrastructure
- AI Gateway: Centralize authentication, auditing, and cost controls for AI tooling
- Administration: Learn how to operate Coder
- Premium: Learn about paid features built for large teams
- IDEs: Connect your existing editor to a workspace
Support
Open an issue for bugs and feature requests.
For community support, showcasing what you built, and feedback on in-progress features, join our Discord or GitHub Discussions.
Dedicated support is included in Coder Premium. For more information, visit coder.com/pricing.
Integrations
New integrations are always in progress. Open an issue to request one. Contributions are welcome in any official or community repository.
Official
- Coder Registry: Templates, modules, and integrations for common development environments
- Coding Agents: Run agents like Claude Code, Codex, and OpenCode isolated in Coder workspaces
- coderd Terraform Provider: Declaratively manage your Coder deployment configuration as code
- VS Code Extension: Open any Coder workspace in VS Code with a single click
- JetBrains Toolbox Plugin: Open any Coder workspace from JetBrains Toolbox with a single click
- Dev Containers: Build development environments using
devcontainer.jsonon Docker, Kubernetes, and OpenShift - Kubernetes Log Stream: Stream Kubernetes Pod events to the Coder startup logs
- Self-Hosted VS Code Extension Marketplace: A private extension marketplace that works in restricted or airgapped networks integrating with code-server.
- GitHub Actions: An action to set up the Coder CLI in GitHub workflows
Community
- Coder Discord: Chat with the community, get support, and hear about new product updates first
- Community Templates: Community-contributed workspace templates in the Coder Registry
- Community Modules: Community-contributed modules to extend Coder templates
- Coder Template GitHub Action: A GitHub Action that updates Coder templates
- Coder Agents Chat Action: A GitHub Action that starts a Coder Agents chat against a GitHub issue or pull request
Contributing
New contributors are always welcome. If you are new to the Coder codebase, see the contribution guide to get started.
Hiring
Apply on the careers page if you are interested in joining the team.