The Atlas Lavern's documentation, bound to its code
111 documents

The Atlas

Lavern's documentation, bound to its code

Journeys

First hour with the repo

The orientation path: what Lavern is, how to run it, where everything lives, and how to navigate from here.

5 stops →

How the engine works

From the prescriptive architecture to the actual dispatch loop: the 4-stage pipeline as designed, then the code that runs engagements.

6 stops →

Clawern deep-dive

The autonomous pipeline — watch a folder, read what arrives, remember across documents — and the seven-report eval arc that hardened it.

8 stops →

Integrating & operating

The system's edges: providers and tools, the API surface, the remote bridge, datasets, and the two live migration roadmaps.

6 stops →

Getting Started

The front door. What Lavern is, how to get it running in a minute, and the personality it speaks with. Start here, in order: README → QUICKSTART → SOUL.

  • Lavern Your very first stop in this repository.
  • Lavern's Soul You're tuning agent tone or customizing the firm personality.
  • Quick Start You want it running on your machine in the next five minutes.

Architecture & Design

Why the system is shaped this way: the 4-stage pipeline, the 25-role agent taxonomy, and the complete codebase topology. The prescriptive specs predate the code; CLAUDE.md tracks what was actually built.

Integration

Everything that touches the outside world: the three LLM providers, 21 MCP tools, bundled legal datasets, external services — plus the two live migration roadmaps (Anthropic Managed Agents, Azure).

Audits & Quality

Point-in-time health reports: a tech-debt audit that found an unusually clean codebase (and the CLAUDE.md drift this Atlas now checks continuously), and the pre-launch security audit behind the fresh-start v0.15.0 repo.

Evaluation

The v1→v3.4 Lighthouse eval arc: seven reports on real CUAD JV contracts running on a local gemma2:2b, each diagnosing a failure mode, fixing it, and re-measuring — from "plumbing works" to zero leaks across every category. RUBRIC.md is the pre-registered baseline; read the arc in order.

Operations

Running and showing the thing: importing the five legal datasets, the demo walkthrough script, the macOS menu bar companion, sample documents, and the bundled legal template library.

Governance

How the project is run: contribution rules built around "every agent must be challengeable", the security policy and trust boundaries, attribution, the code of conduct, and the release history.

  • Authors Provenance or attribution questions.
  • Changelog Upgrade planning, or archaeology on why something changed.
  • Code of Conduct Community conduct expectations.
  • Contributing to Lavern Before your first PR, or when adding an agent or tool.
  • Security You're reporting a vulnerability or assessing the threat model.

Developer Tools

The documentation tools themselves: the Skills & Prompts Explorer (87 artifacts, curated diagrams) and this Atlas. Both follow the same pattern — mechanical discovery strictly separated from authored curation.

Agents & Prompts

All 67 agent prompts, searchable here at full text. For the curated experience — assessment cards, workflow diagrams, click-to-source — each page deep-links into the Explorer.

Marketing Site

The public lavern.ai pages (dark cinematic single-file HTML). Indexed for search; they open as standalone pages, not inside the Atlas.