Summary

LACP is a control-plane-grade agent harness designed for LLMs like Claude, Codex, and Hermes. It provides policy-gated operations, verification/evidence loops, multi-layer memory, and auditable execution, all with a local-first, zero-external-dependency approach. This framework benefits developers seeking predictable, auditable, and secure agent workflows by offering granular control over LLM interactions and outputs.

How It Works

LACP functions as an agent harness with integrated control-plane governance. It enforces policy gates—including risk tiers, budget ceilings, and context contracts—before executing agent tasks, ensuring auditable and compliant outcomes. The system features a 5-layer memory architecture, encompassing session memory, an Obsidian knowledge graph, an ingestion pipeline, code intelligence, and agent identity with provenance. A modular Python hook pipeline further refines agent behavior through session context injection, pre-tool guards, and quality gates.

Quick Start & Requirements

• Installation: Recommended via Homebrew (brew tap 0xNyk/lacp && brew install lacp) or cURL bootstrap (curl -fsSL https://raw.githubusercontent.com/0xNyk/lacp/main/install.sh | bash).
• Prerequisites: bash, python3, jq, rg (ripgrep). shellcheck, tmux, gh are recommended. macOS installer handles missing dependencies via Homebrew.
• Setup: Initial setup and verification can be performed with lacp bootstrap-system --profile starter --with-verify.
• Documentation: Links to Runbook, Local Dev Loop, and other guides are available within the README.

Highlighted Details

• Policy-Gated Execution: Commands are routed through risk tiers, budget ceilings, and context contracts for enhanced security and control.
• 5-Layer Memory Stack: Integrates session memory, Obsidian knowledge graph, ingestion pipeline, code intelligence, and agent identity with hash-chained provenance.
• Modular Hook Pipeline: Python hooks manage session context, guard against dangerous operations, validate writes, and implement quality gates.
• Obsidian Brain Integration: Features first-class vault management and biologically-inspired memory consolidation.
• Multi-Agent Orchestration: Supports dmux/tmux session management, Git worktree isolation, and swarm workflows.
• Local-First Security: Emphasizes zero external CI by default and environment-driven credentials.
• Evidence Pipelines: Generates machine-verifiable evidence for PR gates via browser, API, and smart-contract e2e harnesses.

Maintenance & Community

The project is actively developed by 0xNyk. Community support links like Discord or Slack are not explicitly listed. Professional services related to agent infrastructure are offered by Builderz, with the author reachable at @nyk_builderz.

Licensing & Compatibility

Licensed under the MIT license, which is permissive and generally compatible with commercial use and closed-source linking.

Limitations & Caveats

The project is in an alpha state (v0.9.0) and under active development, meaning APIs and CLI interfaces are subject to change between releases.