Skip to main content

Briefcase AI Documentation

Briefcase AI helps engineering teams build AI workflows that are explainable, reproducible, and operationally scalable in regulated environments.

Why This Matters Now​

AI decision volume is increasing faster than manual review capacity. Teams need a technical way to answer governance questions without reconstructing evidence from fragmented logs.

How It Works​

Briefcase AI combines two capabilities:

  • Versioned decision context for prompts, policies, data references, and configuration artifacts.
  • Runtime decision observability for inputs, outputs, confidence, and execution traces.

Together, these provide replayable, audit-ready decision evidence.

Three Audit Questions Your System Should Answer​

  1. Which versioned artifacts were in effect when this decision ran?
  2. What inputs, outputs, and confidence signals produced this outcome?
  3. Can we replay this decision and reproduce behavior for investigation?

How Engineers Use It​

A common engineering flow is:

  1. Instrument workflow steps with briefcase_ai.
  2. Capture decision traces with version/context metadata.
  3. Query traces through SDK or API.
  4. Replay historical decisions for debugging and verification.
  5. Review low-confidence or policy-sensitive outcomes.

Core Capabilities​

Decision Tracking and Replay​

Capture decision inputs/outputs and replay behavior deterministically.

Versioned Knowledge and Artifact Context​

Track commit-linked knowledge references with decorators, context managers, or direct clients.

Validation and Compliance-Oriented Checks​

Validate prompt-to-knowledge references and produce structured evidence artifacts.

Multi-Step Workflow Correlation​

Propagate workflow context across agent chains and services.

Quick Start​

import briefcase_ai

briefcase_ai.init()

decision = briefcase_ai.DecisionSnapshot("chat_completion")
decision.add_input(briefcase_ai.Input("prompt", "Summarize this text", "string"))
decision.add_output(briefcase_ai.Output("response", "Summary output", "string"))

db = briefcase_ai.SqliteBackend.in_memory()
decision_id = db.save_decision(decision)
print(db.load_decision(decision_id).function_name)

Compatibility aliases remain available directly from briefcase_ai:

from briefcase_ai import versioned_context, briefcase_workflow

The legacy briefcase namespace remains as a compatibility alias in 2.1.30 and is scheduled for removal in 2.1.31.

Installation​

pip install briefcase-ai

Documentation Sections​

LLM Context File​

For AI assistants and development tools, this documentation is available as an optimized context file:

llm.txt - Machine-readable documentation optimized for LLM consumption

Use this URL when asking AI assistants for help with Briefcase AI integration.

Support​

License​

GPL-3.0. See the repository LICENSE file.