Skip to content

Architecture

Design goal

Thin tool layer over the typed @johannes.latzel/seq-client. The package adds no HTTP logic; it turns model-facing tool arguments into seq-client calls and formats the results for the LLM.

Tool layer

Each tool extends the llm-chat Tool base class and takes a SeqClient from @johannes.latzel/seq-client. onExecute validates the JSON-serializable arguments defensively, maps them to the seq-client option surface, and formats the result.

Configuration

SeqConfiguration resolves defaults from LLM_CHAT_SEQ_* environment variables, with constructor arguments taking precedence. createSeqClient builds a SeqClient, throwing a clear error when no url is configured.

Integration tests

The tests/integration/ suite exercises the real HTTP path against an in-process mock Seq server, so npm run test:integration needs no running Seq instance. Set SEQ_TEST_URL to run against a real server instead. The suite is excluded from npm run verify.