canvas-bot
8e360bcc70
The user explicitly asked for the agent to be backed by a fork of the
pi coding agent. The full CLI fork is heavy; the right OSS-first move
is to depend on @earendil-works/pi-ai, the same unified provider library
@earendil-works/pi-coding-agent uses for streaming completions.
What this commit does:
- llm-proxy/server.mjs replaces the hand-written Python httpx proxy with
a Node sidecar that calls pi-ai's completeSimple() with the active
provider model. Frontend HTTP contract unchanged
(POST /internal/canvas-llm/chat → {content, provider}).
- Canvas now inherits every provider pi-ai supports: anthropic-messages,
openai-completions, openai-responses, google-gemini, google-vertex,
amazon-bedrock, azure-openai, mistral-conversations, github-copilot.
- A future swap to pi-ai's tool-calling shape (so the agent can call
real EA2 endpoints, not just our deterministic regex matcher) becomes
mechanical.
- Health endpoint surfaces which providers have keys configured.
- 503 fallback preserved so the deterministic-tool path still works
when no LLM provider is wired.
Image published: gitea.flow-master.ai/shad/canvas-llm-proxy:sha-pi-ai-0.2.0
Digest: sha256:9bc3f3896d2f1346bf210392e0885d5e2a3be77260fafa48be61e4f61c97b085
Old Python sidecar (main.py, requirements.txt, Dockerfile) retained
for one release so deployers can roll back.
canvas-llm-proxy
In-cluster sidecar for canvas.flow-master.ai. Brokers
POST /internal/canvas-llm/chat to the configured upstream LLM provider so
the in-app Command Assistant can synthesise fluent natural-language replies
when no deterministic tool matches.
Why a Node sidecar (not a Python one)
This was a Python httpx proxy that hand-wrote OpenAI and Anthropic
calls. PR canvas-frontend#6 swapped it for a Node sidecar that depends on
@earendil-works/pi-ai
— the same unified provider client @earendil-works/pi-coding-agent uses.
Why:
- It is the OSS-first move the canvas user asked for ("you could use a fork pi coding agent for that agent and then integrate it because pi coding agent is very clean and small while being efficient and useful").
pi-ainormalises streaming completions across OpenAI (completions + responses), Anthropic, Google Gemini, Google Vertex, Amazon Bedrock, Azure OpenAI, Mistral, GitHub Copilot. Canvas inherits all of them.- A future move to pi-ai's tool-calling shape (so the agent can hit real EA2 endpoints, not just our deterministic regex matcher) becomes mechanical instead of a rewrite.
Wire-compatible
The HTTP contract is the same: POST /internal/canvas-llm/chat with
{messages, max_tokens} returns {content, provider}. The frontend's
src/lib/llmClient.ts does not change.
Configuration
Set one of: ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY.
Optional: ANTHROPIC_MODEL, OPENAI_MODEL, GEMINI_MODEL, PORT (8080).
If no key is set, every request returns 503 so the frontend's
deterministic-tool fallback fires gracefully.
Run
cd llm-proxy
npm install
ANTHROPIC_API_KEY=… node server.mjs
Container:
docker build -f Dockerfile.node -t canvas-llm-proxy .
docker run -e ANTHROPIC_API_KEY=… -p 8080:8080 canvas-llm-proxy
Legacy
The previous Python sidecar (main.py, requirements.txt, Dockerfile)
is retained for one release so deployers can roll back. It will be
removed once the Node sidecar is in production.