# Briefcase AI

> Infrastructure for governing the decisions your AI systems make: enforce controls before an action runs, capture the full context behind every decision, and keep a complete, reproducible record you can verify later. Open-source Python SDK with a Rust core.

Briefcase wraps the decision points in your application. A decision moves through five acts — Capture, Control & Route, Store & Query, Replay & Verify, Prove — and these docs are organized the same way. Install with `pip install briefcase-ai`.

A full-text version of every page is at https://briefcaseai.io/llms-full.txt

## Start Here

- [Why Briefcase](https://briefcaseai.io/getting-started/why-briefcase/): Briefcase is infrastructure for governing the decisions your AI systems make — enforce controls before an action runs, capture the full context behind every decision, and keep a complete, reproducible record you can verify later.
- [AI-Assisted Setup](https://briefcaseai.io/getting-started/ai-assisted-setup/): The fastest way to add Briefcase to a project — let your AI coding assistant do it. Point it at the docs, or give it Briefcase's tools over MCP.
- [Quickstart](https://briefcaseai.io/getting-started/quickstart/): Record, persist, and replay your first decision in about 5 minutes.
- [Core Concepts](https://briefcaseai.io/getting-started/core-concepts/): The mental model behind Briefcase — why a decision, its inputs and outputs, its environment, and its grouping are recorded as distinct, reproducible records.
- [Installation](https://briefcaseai.io/getting-started/installation/): Install Briefcase AI, then add only the extras you need — grouped by what you're trying to do, with no surprise dependencies.
- [What's New](https://briefcaseai.io/getting-started/whats-new/): The 3.3.0 release launches oci-bai — an artifact graph for models and fine-tunes — plus the 3.2.x cost rate cards, prompt-cache billing, and latest model pricing.

## Capture

- [Decision Recording](https://briefcaseai.io/features/decision-recording/): Capture every AI decision — inputs, outputs, model, and a content fingerprint — so it can be replayed, audited, and proven later.
- [Exporters](https://briefcaseai.io/features/exporters/): Stream captured decisions to the console, a file, memory, or your own external sink as they happen.
- [PII Sanitization](https://briefcaseai.io/features/pii-sanitization/): Minimize sensitive data — detect and redact it before a decision record is ever stored.

## Control & Route

- [Guardrails](https://briefcaseai.io/advanced/guardrails/): Enforce deny-by-default, fail-closed controls before an agent action runs.
- [Routing](https://briefcaseai.io/advanced/routing/): Route an AI decision between automatic handling and human review with a narrow, in-process gate.
- [Versioned Routing Policy](https://briefcaseai.io/advanced/versioned-routing-policy/): Reconstruct exactly which rule fired and which policy version was active on any past date.
- [Validation Engine](https://briefcaseai.io/advanced/validation-engine/): Block prompts whose references no longer resolve, before they reach a model.

## Store & Query

- [Storage Adapters](https://briefcaseai.io/features/storage-adapters/): Choose where your decision records live — the backend that holds your audit trail.
- [Bitemporal Storage](https://briefcaseai.io/advanced/bitemporal-storage/): Answer "what did we know at decision time?" even after a value is later corrected.
- [External Data](https://briefcaseai.io/advanced/external-data/): Pin the exact upstream value a decision used, so it stays reproducible when the source changes.
- [RAG Versioning](https://briefcaseai.io/advanced/rag-versioning/): Tie each retrieval to the exact corpus state, and catch a stale index before it informs a decision.

## Replay & Verify

- [Deterministic Replay](https://briefcaseai.io/features/replay/): Re-run a saved decision to prove it still produces the same output — and catch the run that doesn't.
- [Drift Detection](https://briefcaseai.io/features/drift-detection/): Catch a model whose answers are quietly getting less consistent over time — before it changes what your system decides.
- [Audit Bundles](https://briefcaseai.io/advanced/compliance-bundles/): Seal a decision, its evidence, and the policy behind it into one portable, tamper-evident artifact.

## Operate

- [Cost Tracking](https://briefcaseai.io/features/cost-tracking/): Estimate token costs across platforms and tiers, bill prompt caching, compare models, project monthly spend, and watch budgets — from the data your decisions already carry.
- [OpenTelemetry](https://briefcaseai.io/features/opentelemetry/): Put governed decisions on the same trace timeline as the rest of your stack.
- [Multi-Agent & Events](https://briefcaseai.io/features/multi-agent/): Tie every decision in a multi-step agent pipeline to one workflow and react to them as they happen.

## Artifact Graph & Evaluate

- [Overview](https://briefcaseai.io/evaluate/runs/): How oci-bai tracks every model, fine-tune, and dataset you push — the commit model, provenance, search, and how to compare versions.
- [Quick Start](https://briefcaseai.io/evaluate/quickstart/): Push your first tracked image, explore the commit graph, and run your first search in under five minutes.
- [CLI Reference](https://briefcaseai.io/evaluate/cli/): Every oci-bai command and its flags.
- [How It Works](https://briefcaseai.io/evaluate/architecture/): The gateway, artifact graph, and how pushes become searchable, versioned, deduplicated commits.

## Guides

- [Audit a Decision End-to-End](https://briefcaseai.io/guides/audit-a-decision/): Follow one support-triage decision from the evidence it depended on, through the policy that governed it, to a sealed, tamper-evident record you can verify months later.
- [Govern Agent Actions](https://briefcaseai.io/guides/govern-agent-actions/): Put controls in front of an agent so an action is evaluated before it runs — deny-by-default guardrails, fail-closed pipelines, and a versioned policy that decides where the work goes.
- [Observe AI in Production](https://briefcaseai.io/guides/observe-in-production/): Wire up live observability for a running triage agent — emit every decision, watch token cost against a budget, measure output drift, trace with OpenTelemetry, and fire events when something looks off.
- [Reproducible RAG](https://briefcaseai.io/guides/reproducible-rag/): Make a retrieval-augmented decision reproducible — version the embedding index, snapshot the external data it read, and validate that a prompt's references resolve before a model ever sees them.
- [Track & Compare Model Versions](https://briefcaseai.io/guides/run-an-evaluation/): Push a baseline and candidate through the gateway, compare them at every depth, and read the verdict scorecard.

## Integrations

- [MCP Server](https://briefcaseai.io/integrations/mcp/): Expose Briefcase's safe, read-only SDK operations to MCP-capable tools.
- [lakeFS](https://briefcaseai.io/integrations/lakefs/): Capture lakeFS commit SHAs for the versioned data your agents read.

## Reference

- [oci-bai Install & Compatibility](https://briefcaseai.io/reference/install-compat/): Install the oci-bai CLI, bring up the local stack, and check the compatibility matrix.
- [Python SDK](https://briefcaseai.io/sdk/python/): Complete guide to the Briefcase AI Python package.
- [Python API](https://briefcaseai.io/api/python/): Full API reference for the Briefcase AI Python SDK, grouped by module.
- [Glossary](https://briefcaseai.io/reference/glossary/): Definitions for the core vocabulary used across Briefcase — decisions, evidence, policies, bitemporal time, and verifiable bundles.

## Resources

- [Architecture](https://briefcaseai.io/resources/architecture/): System architecture of Briefcase AI.
- [Examples](https://briefcaseai.io/resources/examples/): Example code and use cases for Briefcase AI.
- [Changelog](https://briefcaseai.io/resources/changelog/): Release history for Briefcase AI.
- [FAQ](https://briefcaseai.io/resources/faq/): Frequently asked questions about Briefcase AI.

## Contributing

- [Development](https://briefcaseai.io/contributing/development/): Set up a development environment for Briefcase AI.
- [Build the Rust Core](https://briefcaseai.io/sdk/rust/): For contributors — build the Briefcase AI Rust core and Python bindings from source.
- [Testing](https://briefcaseai.io/contributing/testing/): Run and write tests for Briefcase AI.
- [Code Standards](https://briefcaseai.io/contributing/code-standards/): Code style and conventions for Briefcase AI.
- [Governance](https://briefcaseai.io/contributing/governance/): Project governance and decision-making.