typescript-openai-agent

TypeScript OpenAI Agent

OpenAI SDK assistant, bindufied via the TypeScript SDK. The TS SDK spawns the Python bindu core in the background over gRPC — you write bindufy(config, handler) in TypeScript, the agent comes online with a DID, an A2A endpoint, and OAuth, same as a Python agent.

This example points the OpenAI SDK at OpenRouter (baseURL: https://openrouter.ai/api/v1) so it runs on the same key as the rest of the example fleet.

Setup

export OPENROUTER_API_KEY=<get one at https://openrouter.ai/keys>
cd examples/typescript-openai-agent
npm install

Run

npx tsx index.ts
# http://localhost:3773

The TS SDK launches uv run bindu serve --grpc --grpc-port 4774 as a child. If port 4774 is taken (the gateway dev server lives there in dev), change coreAddress in the bindufy config to a free port.

Talk to it

With AUTH__ENABLED=false:

curl -sS http://localhost:3773/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"message/send","id":"00000000-0000-0000-0000-000000000004","params":{"message":{"role":"user","parts":[{"kind":"text","text":"Reply with one plain-text sentence: capital of France?"}],"kind":"message","messageId":"00000000-0000-0000-0000-000000000001","contextId":"00000000-0000-0000-0000-000000000002","taskId":"00000000-0000-0000-0000-000000000003"},"configuration":{"acceptedOutputModes":["application/json"]}}}'

If you ask a vague open-ended question, openai/gpt-4o-mini sometimes replies in JSON with {"state":"input-required","prompt":"..."} — that's not a bug, it's the model asking for clarification, parsed by bindu's response detector into a valid input-required task state. Phrase prompts plainly to get a completed state.

With auth on, sign each body with the agent's DID key — see docs/AUTH.md.