Runnable, self-contained demos and setups for Praxis.
Each demo lives under demos/<name>/ with its own README and bring-up script.
| Demo | Description |
|---|---|
| anthropic-messages | Route Anthropic /v1/messages requests to any backend — Anthropic API, vLLM, or OpenAI-compatible — with optional format transformation via composable filters. |
| cpex-policy-engine | Policy enforcement on MCP traffic: authorization flows connecting identity to access control decisions with Cedar or CEL PDP, delegation, data redaction, and session tainting. |
| openai-responses-stateless | Stateless passthrough for OpenAI /v1/responses with store: false. Praxis classifies the request, detects stateless mode, and proxies directly to vLLM — no buffering, no persistence, no transformation. |
| openai-responses-codex-passthrough | Live Codex CLI passthrough to the OpenAI Responses API. Demonstrates model alias rewriting, default injection, effective-model headers, SSE, and a Codex-owned tool loop. Run the all-in-one narrated demo or each step individually. |
| openai-responses-multi-turn | Multi-turn conversation (non-streaming) for the OpenAI Responses API. Praxis stores turn 1 in SQLite, then rehydrates the conversation history on turn 2 via previous_response_id and rebuilds the request body before forwarding to vLLM. |
demos/
<name>/
README.md # what it shows and how to run it
... # configs, scripts, and any services
Each demo is independent. Start from its README.