How ilirium/ilirium_llm_router works
ilirium_llm_router is an Anthropic-compatible proxy that lets Claude Code use Anthropic and LM Studio in one session. The main workflow is a client request entering the FastAPI app, being forwarded by the proxy to the model-selected backend, and returning as a streamed response. Calls also feed telemetry and, when enabled, an opaque body corpus. CLI commands configure and serve the router, train compression dictionaries, and extract or verify archived calls. Edges reflect README statements and shown source relationships; unsampled implementation details are not inferred.
Main entry points: proxy.py Request proxy · cli.py CLI commands · app.py HTTP application · routing.py Model routing
Components 16
Entry and configuration
- CLI commands
src/ilirium_llm_router/cli.py - Configuration
src/ilirium_llm_router/config.py - HTTP application
src/ilirium_llm_router/app.py
Request routing
- Request proxy
src/ilirium_llm_router/proxy.py - Model routing
src/ilirium_llm_router/routing.py
Call observability
- Usage measurement
src/ilirium_llm_router/observe.py - Telemetry CSV
src/ilirium_llm_router/stats.py - Rotating logs
src/ilirium_llm_router/logging_setup.py
Corpus and replay
- Body corpus
src/ilirium_llm_router/corpus.py - Dictionary training
src/ilirium_llm_router/dictionary.py - Archive extraction
src/ilirium_llm_router/extract.py - JSONL output
src/ilirium_llm_router/jsonl.py - Transcript replay
src/ilirium_llm_router/transcript.py
Other components
- Claude Code
- Anthropic API
- LM Studio
Connections 23
- Claude Code → HTTP application: sends requests
README.md - HTTP application → Request proxy: dispatches routes
src/ilirium_llm_router/app.py - Request proxy → Model routing: selects backend
src/ilirium_llm_router/proxy.py - Model routing → Anthropic API: selects claude models
README.md - Model routing → LM Studio: selects other models
README.md - Request proxy → Anthropic API: forwards requests
README.md - Request proxy → LM Studio: forwards requests
README.md - Anthropic API → Request proxy: streams replies
README.md - LM Studio → Request proxy: streams replies
README.md - Request proxy → Usage measurement: measures calls
src/ilirium_llm_router/proxy.py - Usage measurement → Telemetry CSV: records metrics
src/ilirium_llm_router/observe.py - Request proxy → Telemetry CSV: writes call rows
src/ilirium_llm_router/app.py - HTTP application → Rotating logs: sets up logging
src/ilirium_llm_router/cli.py - Request proxy → Body corpus: captures bodies
src/ilirium_llm_router/proxy.py - Body corpus → Dictionary training: supplies training data
src/ilirium_llm_router/dictionary.py - CLI commands → Configuration: loads settings
src/ilirium_llm_router/cli.py - CLI commands → HTTP application: starts server
src/ilirium_llm_router/cli.py - CLI commands → Dictionary training: runs training
src/ilirium_llm_router/cli.py - CLI commands → Archive extraction: selects archived calls
src/ilirium_llm_router/cli.py - Archive extraction → JSONL output: writes JSONL
src/ilirium_llm_router/extract.py - Archive extraction → Transcript replay: replays replies
src/ilirium_llm_router/extract.py - JSONL output → Transcript replay: uses replay logic
src/ilirium_llm_router/jsonl.py - CLI commands → Body corpus: reads archive
src/ilirium_llm_router/cli.py