KimGLee/Cambium

Governance standard and reference toolset for LLM-maintained knowledge corpora

What it solves

Cambium provides a governance standard and toolset for knowledge repositories that are maintained by LLM agents. It addresses the lack of structured rules and evidence-based tracking for agentic work, ensuring that agents do not guess when resuming tasks, know exactly which rules apply to a repository, and maintain a clear boundary between operator decisions and agent actions.

How it works

Cambium operates as a governance layer that sits between the agent host and the knowledge repository. It uses a "kernel" of normative rules, a selected "profile" that defines the repository's specific scope and priorities, and a runtime state (ledgers) to track progress.

Key components include:

  • The Kernel: Defines cross-domain governance semantics and invariants.
  • Profiles: Repository-specific configurations (TOML) that define roles, architecture, and priorities.
  • Ledgers: Three state objects (Coverage, Required Queue, and Progress) that track which objects exist, what work is pending, and the current task contract.
  • Tools: A set of deterministic CLI commands for controlled writes, schema checks, and state transitions.
  • MCP Projection: A host-neutral interface that allows agent hosts (like Claude Code or Kimi Code) to interact with Cambium's tools via the Model Context Protocol.

Who it’s for

It is designed for operators who use LLM agents to maintain large-scale knowledge bases or technical repositories where strict adherence to rules, resumability of long-running tasks, and verifiable evidence of completion are required.

Highlights

  • Resumable Work: Uses persistent state and ledgers to allow tasks to be interrupted and resumed without loss of context or guessing.
  • Deterministic Governance: Tools execute declared rules rather than relying on the agent's semantic judgment.
  • Agent-Host Integration: Provides generated configurations for multiple agent hosts (Claude Code, Codex, Kimi Code, dsh) via MCP.
  • Evidence-Based Closure: Requires append-only receipts and terminal proof bindings before work can be closed.
  • Controlled State Changes: Shared state is modified only through specific writer tools to ensure history and hashes remain consistent.

Related

  • Project
    nvk/llm-wiki

    A knowledge management framework that enables AI agents to research, compile, and maintain structured wikis for personal projects and complex information gathering.

  • Project
    cordum-io/cordum

    Cordum is an agent control plane that provides a deterministic governance layer to intercept, audit, and approve autonomous AI agent actions before they are executed.

  • Project
    mnemon-dev/mnemon

    Mnemon is a cross‑platform CLI that adds persistent, graph‑based memory to LLM agents. The host LLM supervises what to remember, link, or forget, while Mnemon handles deterministic storage, decay, deduplication, and optional vector search. It integrates with many AI coding tools (Claude Code, Codex, Cursor, etc.) via one‑click setup, requiring only a single local binary and no extra API keys.

  • Project
    TokenRollAI/llmdoc

    A persistent engineering context system that helps coding agents understand a repository's architecture and design decisions without rediscovering them every session.

  • Project
    henryqin1997/statem

    A command-line state machine that turns AI agent workflows into inspectable graphs of states and executable checks to ensure reliability in long-running tasks.