Clawpy DocsMain Site

Welcome to Clawpy

Clawpy is an open-source autonomous multi-agent orchestration platform. It coordinates specialized LLM swarms inside secure, isolated Docker-out-of-Docker (DooD) sandboxes—delivering deep task execution, transparent tri-layer memory, and production-grade safety controls.

What is Clawpy?

Unlike single-model chat assistants or generic agent frameworks, Clawpy runs a highly coordinated swarm of permanent, purpose-built Matrix agents that collaborate to complete complex, multi-step engineering and business operations. Each agent operates with its own distinct identity, dynamic tri-layer memory, and strict sandboxed workspace—all observable and configurable through a beautiful, unified developer dashboard.

You describe what you need in natural language. Clawpy automatically delegates the work to the right specialists, monitors execution, manages budgets, and delivers validated outcomes—while preserving a complete audit trail of every decision and code change.

Core Concepts

  • Specialized Swarms — A collaborative team of purpose-built AI specialists. Swarms are dynamically configured and routed via the Stack Profile Configurator to ensure tasks are assigned to the most efficient expert models without bottlenecks.
  • 7 Matrix Agent Identities — Reliable task execution is guaranteed by mapping all workloads to seven permanent, expert archetypes defined in the core templates:
    • Alfred (Chief of Staff / Computer Operator): The conversational entry point and personal operator aide who coordinates calendar, workspace setup, and delegation.
    • Lucius (CFO / COO Business Operations): Leads business intelligence, budgeting, transactional marketplace kits, growth strategies, and growth reviews.
    • Guardian (CISO / Safety Sentinel): Scans intent ciphers, guards prompt-injection boundaries, and enforces strict action approval gates.
    • Morpheus (CTO / Engineering Lead): Architect of systems, developer of code, and commander of full-stack test execution and DevOps deployment.
    • Trinity (CMO / Creative Director): Oversees audience-facing brand aesthetics, media production, content structure, sitemaps, and SEO strategies.
    • Oracle (Research Director): Drives deep market intelligence, data analysis, competitive studies, and long-horizon knowledge discovery.
    • Neo (QA Lead / Compliance Reviewer): Orchestrates structured multi-agent debates, release gates, and strict verification checklists before code is written.
  • Tri-Layer Memory & Ebbinghaus Decay — A persistent memory architecture operating across three synchronized layers: Markdown (Obsidian-ready) for human readability, SQLite (FTS5) for fast keyword indexing, and ChromaDB for semantic vector retrieval. Retained context is prioritized via Maximal Marginal Relevance (MMR) and optimized dynamically using a 4-tier Ebbinghaus Forgetting Curve model (Goldfish, Standard, Elephant, Permanent) where active memory access extends context lifespan.
  • Proactive Heartbeat Loop — An autonomous background daemon running persistent cron routines 24/7 via APScheduler. This allows agent swarms to wake up on scheduled timers, check system health, perform recurring routines, and execute background operations without requiring constant operator prompts.
  • Strict DooD Sandboxing — Hardware-grade execution safety. Clawpy isolates all shell operations and code execution inside strict Docker-out-of-Docker (DooD) quarantine containers with zero access to your host filesystem or network unless explicitly authorized by the operator.

How It Works

  1. Describe what you need — Explain your operational goals in plain language inside the operator console or via integrations (Telegram/Discord).
  2. Stack configuration & routing — The Stack Profile Configurator determines the optimal agent mix, sets model boundaries, and assigns workloads.
  3. The swarm executes — Specialized Matrix agents collaborate, run multi-agent debates to resolve ambiguity, write tests, and execute isolated sandbox code.
  4. Review the output & logs — Inspect the results, review complete step-by-step decision trees, and explore updated temporal memory graphs.
  5. Iterate or automate — Refine outcomes conversationally or schedule the workflow to run as a recurring, hands-free routine in the background.

Open Source vs Managed

Clawpy is fully open-source. You can self-host the platform on any local system or Linux VPS with complete control over your private data, custom agent profiles, and model API keys.

Clawpy Managed (coming soon) will offer a production-ready cloud dashboard with one-click multi-vps deployments, team collaboration interfaces, advanced telemetry metrics, and premium pre-built marketplace kits.

Get Started

Ready to deploy? Head to the Quickstart guide to get your first secure swarm running in under 10 minutes.