An AI-powered governance and workflow CLI for software development teams. Agent automates story creation, implementation planning, code review, and compliance enforcement — all from the command line.

Read our core philosophy on how agentic-dev approaches AI-assisted engineering:

• Engineering Discipline in the Age of AI: The End of "Vibe Coding"
• Commanding the AI Assembly Line: Configuration, Ingestion, and Visualization
• Credential Security in the Agentic Workflow: Safety Meets Absolute Convenience
• Eradicating Developer Toil: The Magic of the Agentic CLI
• Admin and Voice: Visibility and Feedback Loops
• Agentic Console: Real-time AI Programming Interface

• Story & Runbook Management — Create, track, and implement user stories with structured workflows and state transitions.
• AI Governance Panel — Multi-role preflight checks (Security, Architect, QA, Compliance, Observability) that validate your changes before commit.
• Parallel ADK Engine — Blazing fast governance evaluation leveraging the Google Agent Development Kit for concurrent multi-agent analysis.
• Oracle Preflight Pattern — Advanced context retrieval fusing Notion, NotebookLM via MCP, and an embedded zero-server Vector database for high fidelity AI decisions.
• Multi-Provider AI — Works with Gemini, Vertex AI, OpenAI, Anthropic, GitHub Copilot, and Ollama (local). Automatic fallback between providers on rate limits.
• Smart Test Selection — Performs real-time Python impact analysis to intelligently group and selectively execute necessary tests.
• User Journey Testing — Define user journeys as YAML, auto-generate test stubs, enforce implementation gates, and track test coverage.
• Voice UX Reviews — Analyze hands-free voice sessions (agent review-voice) to grade agent latency, accuracy, tone, and interruption handling.
• Automated License Headers — Enforces and automatically generates required copyright headers across specific file types in the project.

See Getting Started for full instructions on prerequisites, how to get .agent into your repository, configuration, and running agent onboard.

The standard Agentic Development workflow follows a strict requirements-to-code pipeline:

# 1. Create a tracking story
agent new-story INFRA-001

# 2. Automatically generate an implementation plan (Runbook)
agent new-runbook INFRA-001

# 3. Have the AI implement the approved Runbook
agent implement INFRA-001

# 4. Run the Parallel Governance Council checks
# (Architect, Security, QA, Compliance, etc.)
agent preflight --story INFRA-001

# 5. Commit with story tracking and automated message
agent commit

We enforce a 500 LOC ceiling per file and circular dependency checks. To run locally:

agent preflight --gate quality
# OR
python3 scripts/check_loc.py
python3 scripts/check_imports.py

The agent review-voice command enables you to evaluate the quality of a voice agent session. It analyzes latency, accuracy, tone, and interruption handling to provide structured UX feedback on the voice bot's performance.

The agent mcp auth command manages authentication for MCP servers (e.g., NotebookLM). See the NotebookLM Integration Guide for full setup and configuration.

• --auto: Automatically extracts session cookies from a supported local browser using the OS-native keychain (requires explicit GDPR consent).
• --file <path>: Uses a file-based cookie import from the provided path instead of launching Chrome interactively.
• --no-auto-launch: Prints instructions for manual cookie extraction instead of launching the interactive browser script.
• --clear-session: Clears the saved authentication session cookies for this server from the secure storage.

The agent sync notebooklm command syncs notebook context into the local database cache using the MCP functionality.

• --reset: Clears internal caching state only, forcing a fresh fetch on the next sync.
• --flush: Clears both internal caching state and deletes all previously tracked artifacts and sources from the database cache.

Ollama enables fully local, self-hosted AI inference with no data leaving your machine. Any model available in Ollama can be used.

# Start Ollama and pull a model
ollama serve && ollama pull llama3

# Use with any agent command
agent query "how does the router work?" --provider ollama

OpenTelemetry instrumentation is available for LLM flows and spans.

To provide robust execution, a standardized @with_retry and retry_async / retry_sync utility is available. By default, it uses exponential backoff and safely interacts with telemetry. If a non-transient error occurs, raise NonRetryableError to immediately abort the retry loop.

• Onboarding Guide — How to install and configure .agent in your repository
• Provider Setup — AI provider comparison, credentials, and advanced configuration
• Release Guide — Packaging and release process
• Configuration & Visualization — Managing the AI assembly line
• Credential Security — AES-256 and native OS keyring integration
• Developer Experience — Frictionless operations via match-story, commit, and pr
• Admin and Voice — Embedded visibility and continuous UX refinement
• NotebookLM Integration — Configuration, authentication, and sync for NotebookLM MCP
• ADRs — Architectural decision records
• Workflows — Detailed workflow instructions

Apache License 2.0

Copyright 2026 Justin Cook