Looma is a powerful tool designed to transform coding-agent history into a coherent, resumable project context. Rather than sifting through transcripts, Looma effectively reconstructs essential insights on:

• Active work
• Key decisions
• Blockers
• Commits
• Files in progress
• Likely next steps

Local-First Approach

Looma operates entirely on your local machine, ensuring privacy and security without the need for cloud services or API keys. This design allows for seamless interaction with coding history while maintaining control over your data.

Demo recorded with a synthetic sample project - see docs/demo/ for details.

Key Features

• Current Version: v2.1.5
• Integrates with services like Claude Code, Codex, and Cursor history.
• Clearly distinguishes between real data and heuristic predictions.

Usage Examples

Looma offers a range of commands to streamline project management and enhance productivity:

looma doctor              # Verify the environment setup
looma ingest              # Index coding-agent history (auto-creates the database)
looma                     # Daily overview: working on, changed, blocked, next steps
looma weekly              # Weekly summary of all repositories: work done, decisions made
looma pack                # Create a minimal context package for another agent
looma inspect             # Analyze repository architecture, systems, and risks
looma resume "auth"      # Reconstruct context for an ongoing task
looma explain "auth"     # Understand the evolution of a WorkItem
looma ask "why postgres"  # Retrieve validated memory and work items

The command looma provides an immediate view of daily tasks and pertinent next steps for various repositories, facilitating context switching with ease.

How It Works

Looma processes coding history to create a structured output that includes:

Claude Code history
      |
      v
Normalized Events (vendor-agnostic - raw_json preserved)
      |
      v
WorkItems (features, bug fixes, refactorings, migrations)
      |
      v
Candidate Memories (decisions, todos, bugs, architecture notes)
      |
      v
Confidence (derived from file overlap, commit linkage, and more)
      |
      v
Resume Bundle (context that is WorkItem-first and git-anchored)

Efficient and Privacy-Conscious

By default, Looma utilizes a fast heuristic extraction method with no external dependencies, ensuring a lightweight installation. For users running local OpenAI-compatible model servers, Looma can automatically integrate these, enhancing extraction precision.

For more technical details and architecture design, refer to the ARCHITECTURE.md.

Distinct from Transcripts

Where conventional transcript searches focus on specific queries, Looma provides a broader context around active work, decisions, and unfinished tasks. This focus on WorkItems enables a more effective organization of historical coding efforts.

Conclusion

Looma not only enhances productivity by providing an organized view of coding-agent history but also prioritizes user privacy with its local-first approach. Designed for developers seeking to gain insights into their projects without compromising on data security.