FAQ

What is Briefcase AI?

Briefcase AI is an open-source Python SDK for recording, replaying, and auditing AI decisions. It captures every input, output, and parameter of AI calls as immutable snapshots.

What languages are supported?

You install and use Briefcase in Python — pip install briefcase-ai, no Rust toolchain required. The performance-critical core is written in Rust and ships precompiled inside the wheel. See Build the Rust Core only if you want to contribute to that core.

Which AI frameworks does it support?

The open-source SDK provides extensible framework protocols (guardrails, exporters, routers, event emitters). Pre-built integrations for LangChain, CrewAI, LlamaIndex, AutoGen, AG2, and OpenAI Agents are available in Briefcase AI Enterprise.

Where are persistent decisions stored?

A DecisionSnapshot you persist goes to a storage backend: by default an in-memory SQLite database (non-persistent), or a SQLite file / custom backend you configure for production. See Storage Adapters.

How does @capture emit decisions?

@capture is a separate, lighter path — it sends to an exporter, not a storage backend. It records a lightweight dict per call and emits it once you call briefcase.observe(...). Pass "console" to write JSON lines to stderr, a path ending in .jsonl to append to a file, or "memory" to collect records in MemoryExporter.records. @capture exports in a background thread by default; use @capture(async_capture=False) to make a record available synchronously.

Is there an MCP server?

Yes. Install the mcp extra (pip install briefcase-ai[mcp]) and run briefcase-mcp (or python -m briefcase.mcp). It exposes sanitize_text, estimate_cost, analyze_drift, and how_to tools to MCP-capable clients. See the Python API Reference.

Does it add latency?

The Rust core adds microseconds of overhead per decision. Storage write latency depends on your backend choice.

Is it production-ready?

Briefcase AI is under active development. Check the Changelog for the latest release status.

How do I contribute?

See Development and Code Standards.