PersonalClaw

A self-hosted AI agent platform where every channel gets its own identity, memory, skills, and tools -- fully managed through a web dashboard. Features a 3-tier memory system (working memory, conversation history with auto-compaction, and semantic long-term recall via pgvector), 4-provider LLM fallback chain, MCP tool integration with per-channel and global scoping, human-in-the-loop approval policies, self-improving skill generation, sub-agent delegation, and per-channel cost tracking. Currently supports Slack with a platform-agnostic adapter layer for expansion.

Inspired by the OpenClaw ecosystem.

Features

• Channel Isolation -- each channel gets independent identity, memory, skills, tools, and config
• 3-Tier Memory -- Valkey working memory, Postgres conversation history with auto-compaction, pgvector semantic recall with hybrid search
• 4-Provider LLM Fallback -- Anthropic, Bedrock, OpenAI, Ollama with automatic failover on rate-limit or timeout
• MCP Tool Integration -- global + per-channel MCP servers via SSE, HTTP, or stdio transports with allow/deny tool policies
• Human-in-the-Loop Safeguards -- two-layer approval system: plan confirmation + per-tool gateway (ask, allowlist, deny, auto policies)
• Web Dashboard -- 8-tab channel management (Identity, Skills, Memory, Conversations, Schedules, Approvals, MCP, Settings), global MCP page, usage/cost dashboard
• Hot-Reload Config -- dashboard changes push to the backend instantly via WebSocket
• Self-Improving Skills -- tracks workflow patterns, auto-generates skill drafts when patterns repeat, effectiveness tracking with user feedback
• Security & Guardrails -- pre/post content filtering, prompt injection detection, sandboxed execution, PII masking, CLI command validation
• Sub-Agent Delegation -- spawn parallel background tasks with isolated context, filtered tools, and configurable models
• Browser Automation -- Playwright-powered screenshots, web scraping, and form filling
• Scheduled Jobs & Heartbeat -- cron-based scheduled prompts with user notifications, periodic proactive monitoring via heartbeat
• Cost Tracking -- per-channel token usage logging, daily spend charts, budget alerts
• Slash Commands -- /pclaw commands for status, model switching, memory, compaction -- zero LLM cost
• Platform Agnostic -- ChannelAdapter abstraction supports Slack today, extensible to Discord, Teams, CLI
• Prompt Composition Modes -- every-turn, once, or minimal injection for cost/accuracy tradeoff

Screenshots

Login

Dashboard Home

Channel — Identity & Personality

Channel — Skills

Channel — MCP Servers

Channel — Memory

Channel — Conversations

Channel — Schedules

Channel — Approval Policies

Channel — Settings

Global MCP Servers

Usage & Costs

Key Design Principles

• Channel isolation -- each channel has independent config, memory, and tools
• Provider agnosticism -- 4-provider fallback chain (Anthropic, Bedrock, OpenAI, Ollama) via Vercel AI SDK
• Memory-first -- 3-tier memory system (Valkey working memory, Postgres conversations, pgvector long-term)
• Dashboard-driven -- all configuration via web UI with hot-reload to backend
• Platform agnosticism -- ChannelAdapter interface decouples agent engine from messaging platforms

For the full tech stack, architecture, and database schema, see docs/ARCHITECTURE.md.

Prerequisites

• Bun >= 1.3
• Docker (for PostgreSQL + Valkey)
• Node.js >= 22

Quick Start

# 1. Clone and install
git clone <repo-url> && cd personal-claw
bun install

# 2. Start infrastructure
bun run docker:up

# 3. Configure environment
cp .env.example .env
# Edit .env with your API keys (see docs/ARCHITECTURE.md for full variable reference)

# 4. Run database migrations
bun run db:migrate

# 5. Start development servers
bun run dev

The dashboard will be at http://localhost:3000 and the API at http://localhost:4000.

Commands

Command	Description
bun run dev	Start all apps in dev mode (Turborepo)
bun run build	Production build
bun run test	Run all tests (Bun test runner)
bun run check	Lint + format check (Biome)
bun run check:fix	Auto-fix lint + format issues
bun run db:generate	Generate Drizzle migrations after schema changes
bun run db:migrate	Apply pending migrations
bun run db:studio	Open Drizzle Studio (DB browser)
bun run docker:up	Start PostgreSQL + Valkey containers
bun run docker:down	Stop containers

Documentation

Document	Description
AGENTS.md	AI agent coding conventions and architecture rules
docs/ARCHITECTURE.md	System architecture, tech stack, schema, data flows
docs/CHANNELS.md	Multi-platform channel adapter guide
docs/SAFEGUARDS.md	Human-in-the-loop approval system
docs/SETUP_GOOGLE_OAUTH.md	Google OAuth setup guide
docs/SETUP_SLACK_BOT.md	Slack bot setup guide

License

Private project.