DOCSEARCHbyAlgolia
Submit Docs

OneManCompany Documentation

repository·main·Indexed 18 days ago

https://github.com/1mancompany/onemancompany

An open-source Agent Operating System that allows users to build and run AI-powered companies from a browser. It orchestrates hierarchical teams of AI agents—including a Founding Team (EA, HR, COO, CSO) and specialized employees from a Talent Market—to deliver complex projects autonomously. The system supports custom LangChain tool integration, OAuth authentication, and a self-improving agent loop for continuous skill evolution.

Tokens
155.9K
Snippets
303
Records
672
Agent score
63%

What's inside OneManCompany

  1. Overview of the Self-Improving Agent

    main

    The self-improving-agent is a universal system designed to learn from all skill experiences. It uses a multi-memory architecture to continuously evolve its own codebase and capabilities. It implements a feedback loop that extracts experiences, abstracts them into patterns, and updates relevant skills.

    Core Capabilities:

    • Multi-Memory Architecture: Uses Semantic, Episodic, and Working memory.
    • Self-Correction: Detects and fixes errors in skill guidance.
    • Self-Validation: Periodically verifies the accuracy of skills.
    • Hooks Integration: Automatically triggers actions on skill events like before_start, after_complete, and on_error.

  3. What is OneManCompany?

    main

    OneManCompany is an open-source AI Operating System (OS) designed to allow anyone to build and operate a fully AI-driven company directly in a browser. Unlike simple AI chatbots or task-oriented tools, it provides a complete organizational infrastructure including hierarchy, recruitment, task management, performance reviews, and knowledge management.

    Key Characteristics:

    • CEO Role: You are the sole human; all other roles (HR, COO, Engineers, Designers) are independent, thinking, and collaborating AI employees.
    • Scalable Infrastructure: It acts as an OS that abstracts AI model differences (e.g., Claude Code vs. OpenClaw) through a unified base layer.
    • Product-Grade Output: Designed to deliver complete projects (games, comics, software) rather than just drafts.
    • Organizational Intelligence: The system learns from every project, accumulating insights into a company-wide knowledge base and refining employee behavior through 1-on-1s and performance reviews.

  4. Explore the OMC Implementation and Architecture

    main

    The technical implementation of OMC consists of several key components:

    • Backend & Frontend: A FastAPI + WebSocket backend paired with a Canvas 2D + G6 frontend.
    • Employee Data Model: Agents (Employees) are defined using profile.yaml, guidance.yaml, and work_principles.md.
    • Executor Backends: Supports various execution modes including LangChainExecutor, ClaudeSessionExecutor, ScriptExecutor, and CeoExecutor.
    • MCP Tool Bridge: An MCP (Model Context Protocol) server used to integrate self-hosted employees.
    • Workflow Engine: Uses markdown-driven definitions to manage workflows.
    • Unified CEO Comms: Uses ConversationService (replacing the older CeoBroker) which includes a pending queue and auto-reply capabilities.
    • Skill Hooks: Provides lifecycle hooks (similar to Cloudflare Workers) for company-hosted agents.
    • Product Management System: A first-principles Agile system managing Product, Issue, Version, and KR (Key Results).

  5. Understand the Operations Management Capability

    main

    The Operations capability is designed to manage the daily functional aspects of the company. It focuses on maintaining smooth daily operations, optimizing workflows, and managing company resources effectively.

    Core Responsibilities

    • Daily Operations: Ensuring smooth day-to-day company activities.
    • Workflow Optimization: Improving efficiency and streamlining processes.
    • Resource Management: Monitoring and managing available company resources.

    Operational Scope

    Operations management covers the following domains:

    • Process Standardization: Creating and maintaining internal process standards.
    • Resource Allocation: Managing scheduling and the distribution of resources.
    • Coordination: Facilitating cross-departmental communication and alignment.
    • Data Analysis: Performing operations-focused data analysis and generating reports.

  6. Use Multi-Channel Communications via OpenClaw gateway

    main

    The Multi-Channel Communications skill allows you to manage conversations across 20+ messaging platforms simultaneously using an OpenClaw gateway. It supports a wide range of channels including WhatsApp, Telegram, Slack, Discord, iMessage (via BlueBubbles), Signal, Google Chat, Microsoft Teams, Matrix, IRC, LINE, Feishu, and Twitch.

    Key capabilities include:

    • Messaging: Send and receive messages across all connected channels.
    • Session Management: Isolated sessions per conversation with persistent state.
    • Voice Interaction: Support for wake words and TTS via ElevenLabs.
    • Automation: Browser automation (managed Chrome), webhooks, and cron jobs for scheduled tasks.
    • Inter-agent Messaging: Routing messages between multiple agent workspaces.

  8. What is the Ontology and how does it work?

    main

    The Ontology is a typed knowledge graph used for structured agent memory and composable skills. It represents knowledge as a verifiable graph where everything is an entity with a type, properties, and relations to other entities.

    Core Data Models

    • Entity: { id, type, properties, relations, created, updated }
    • Relation: { from_id, relation_type, to_id, properties }

    When to use the Ontology

    Use the ontology when you need to:

    • Remember information: Create or update entities (e.g., "Remember that Alice is a manager").
    • Query knowledge: Ask questions about existing data (e.g., "What do I know about Project X?").
    • Link data: Establish relationships (e.g., "Link Task A to Project B").
    • Traverse dependencies: Find related items (e.g., "Show all tasks for Project Z" or "What depends on X?").
    • Plan work: Model multi-step actions as a sequence of graph transformations.
    • Share state: Allow different skills to access and modify the same set of objects.

  9. What is a Talent in OneManCompany

    main

    In the OneManCompany system, an AI employee is composed of two parts: a Vessel and a Talent.

    • Vessel: The execution container provided by the platform. It manages scheduling, retries, and communication.
    • Talent: The capability package you provide. This includes the Agent's identity, system prompt, skill definitions, and tool configuration.

    To deploy an agent, you package it as a Talent and submit it to the Talent Market.

  10. What is the Ontology and when to use it

    main

    The Ontology is a typed knowledge graph used for structured agent memory and composable skills. It represents knowledge as a verifiable graph where everything is an entity with a type, properties, and relations to other entities. Every mutation is validated against type constraints before being committed.

    Use Cases

    • Memory/Knowledge Retrieval: Triggered by prompts like "Remember that..." or "What do I know about X?".
    • Relationship Management: Triggered by "Link X to Y" or "Show dependencies".
    • Complex Planning: Modeling multi-step work as a sequence of graph transformations.
    • Shared State: Allowing different skills to access and modify the same structured data.

  11. Understand CEO Approval and Oversight Logic

    main

    The system is designed for autonomous operation with targeted human intervention. The EA (Executive Assistant) handles routine operations, low-risk internal work, and status reporting autonomously.

    CEO intervention is required when an action is deemed risky, triggering report_to_ceo(action_required=True).

    Mandatory CEO Approvals:

    • Financial: Budgets, purchases, and pricing.
    • Personnel: Hiring, firing, and promotions.
    • External: Client communications and public announcements.
    • Irreversible Actions: Data deletion or production deployments.
    • File Edits: Any changes made outside an employee's immediate workspace (managed via the Resolution system).

  12. How the Self-Improving Agent lifecycle works

    main

    The Self-Improving Agent operates on a continuous loop triggered by skill interactions. When a skill completes (either via automatic hooks or manual triggers), the agent executes the following workflow:

    1. Analyzes: Evaluates the outcome of the recent interaction.
    2. Extracts: Identifies new patterns, insights, or corrections.
    3. Updates: Modifies relevant skill files to incorporate learned knowledge.
    4. Logs: Records the event into the Multi-Memory Architecture (Semantic, Episodic, and Working memory).
    5. Reports: Provides a summary of the learning event to the user.

    This process ensures that the agent evolves its capabilities based on real-world tool usage and user feedback.