# AI Partner > Self-hosted AI agent that executes goals autonomously β€” web search, code execution, document generation, browser automation, delivered to Telegram/Discord/Slack. Docker, 18+ LLMs, 17 MCP servers. v2 adding human proxy + meeting join. - **Type:** MCP server - **Install:** `agentstack add mcp-amitkrpaliwal-ai-partner` - **Verified:** Pending review - **Seller:** [Amitkrpaliwal](https://agentstack.voostack.com/s/amitkrpaliwal) - **Installs:** 0 - **Category:** [Cloud & Infrastructure](https://agentstack.voostack.com/c/cloud-infrastructure) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [Amitkrpaliwal](https://github.com/Amitkrpaliwal) - **Source:** https://github.com/Amitkrpaliwal/AI-Partner - **Website:** https://aipartner.amitpaliwal.com/ ## Install ```sh agentstack add mcp-amitkrpaliwal-ai-partner ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About πŸ€– AI Partner Autonomous multi-agent orchestration platform that executes complex goals end-to-end β€” without hand-holding. Most AI tools work for you. AI Partner works as you. Give it a goal in plain English. It researches, writes code, generates documents, and delivers results to your Telegram, Discord, or Slack β€” autonomously. 🌐 Website Β· Β· [](LICENSE) [](https://github.com/AmitkrPaiwal/AI-Partner/stargazers) [](https://github.com/AmitkrPaiwal/AI-Partner/network/members) [](https://github.com/AmitkrPaiwal/AI-Partner/commits/main) [](https://github.com/AmitkrPaiwal/AI-Partner/issues) [](docker-compose.yml) [](https://www.typescriptlang.org/) [](server/src/tests) --- [**⚑ Quick Start**](#-quick-start) Β· [**✨ Features**](#-features) Β· [**πŸ€– Agent Profiles**](#-agent-profiles) Β· [**πŸ”Œ Integrations**](#-integrations) Β· [**πŸ“ Architecture**](#-architecture) Β· [**πŸ—ΊοΈ Ecosystem**](#-ecosystem-landscape) --- ## 🎬 See It In Action β–Ά Goal β†’ autonomous execution β–Ά Browser automation & delivery --- ## 🎯 What Is AI Partner? AI Partner is a **self-hosted, autonomous AI agent platform** you run on your own machine with Docker. You describe a goal β€” it decomposes it into tasks, executes them using real tools (web search, code execution, file generation, messaging), validates outcomes against measurable criteria, and delivers results to you automatically. ``` "Research the top 10 AI tools launched this week, write a PDF comparison report, and send it to my Telegram." ``` AI Partner will: search the web β†’ extract data β†’ analyse results β†’ generate a PDF β†’ send to Telegram β†’ confirm delivery. **You don't touch it again.** --- ## ⚑ Quick Start > **Requirements:** [Docker Desktop](https://www.docker.com/products/docker-desktop/) + one LLM API key (OpenAI, Anthropic, Gemini, Groq, DeepSeek, or local Ollama) **Mac / Linux β€” one command:** ```bash curl -fsSL https://raw.githubusercontent.com/AmitkrPaiwal/AI-Partner/main/setup.sh | bash ``` **Windows β€” paste into PowerShell:** ```powershell iwr -useb https://raw.githubusercontent.com/AmitkrPaiwal/AI-Partner/main/install.ps1 | iex ``` **Or manually:** ```bash git clone https://github.com/AmitkrPaiwal/AI-Partner cd AI-Partner ./setup.sh # Mac/Linux .\install.ps1 # Windows (PowerShell) ``` The installer walks you through choosing an LLM provider, entering your API key, and opens the browser UI automatically. **First run takes 2–4 minutes** (Docker image build). --- ## ✨ Features ### 🧠 Autonomous Goal Execution Type a goal β€” AI Partner decomposes it, builds an execution plan, runs it with real tools, validates outcomes against measurable success criteria, and **retries or replans on failure**. - βœ… Up to **3 concurrent goals**, each with up to **5 parallel sub-agents** - βœ… ReAct loop: Reason β†’ Act β†’ Assess β†’ Retry - βœ… Self-correcting: if a script errors, it semantically repairs and re-runs - βœ… Typed success criteria β€” agent **proves** completion, doesn't just say "done" ### πŸ€– 16 Specialist Agent Profiles Pre-built agents with enforced tool whitelists, iteration caps, and automatic routing based on keywords. | Cluster | Agents | |---------|--------| | Research | Web Researcher, Fact Checker, Trend Spotter | | Dev | Python Developer, Node.js Developer, Debugger, Shell Operator | | Data | Financial Analyst, Data Analyst, Excel Builder | | Content | Report Generator, Summarizer, Tech Writer, Prompt Architect, Task Planner | | Delivery | Telegram Reporter | Invoke directly: `@fin-analyst what is RELIANCE.NS today?` Or let keywords auto-route: typing "trending AI tools" fires `@trend-spotter` automatically. ### 🌐 Live Browser Automation Puppeteer-powered browser with **live CDP screencasting** visible in the UI. When a CAPTCHA appears, the agent pauses and shows a **"Solve CAPTCHA β€” Take Control"** button. You solve it, the agent resumes. ### πŸ“¬ Goal-Integrated Messaging Delivery Results aren't just saved β€” they're **validated delivery goals**. The agent marks a task failed if `messaging_send_file` doesn't succeed. Supports: **Telegram Β· Discord Β· Slack Β· WhatsApp Β· Signal** ### 🧠 Persistent Memory - **Episodic memory** β€” timestamped event log of every conversation and outcome - **Vector search** β€” semantic similarity across 4 embedding backends - **Persona** β€” biographic facts and preferences injected into every prompt - **Knowledge base** β€” upload PDFs/docs for RAG retrieval ### πŸ“„ Document Generation PDF Β· Excel (xlsx) Β· PowerPoint (pptx) Β· Word (docx) Β· HTML β€” downloadable from the UI or sent via messaging. ### πŸ“š Skill Learning After a successful goal, AI Partner generalises the solution into a **reusable parameterised skill template**. Deduplicated by embedding similarity. Skills can be promoted to first-class MCP tools. ### ⏰ Scheduler + Triggers Cron-expression scheduling, webhook triggers, **Google Calendar events**, **Gmail arrival** β€” all fire autonomous goal execution. --- ## πŸ€– Agent Profiles Each profile specifies: | Field | Description | |-------|-------------| | **Tool whitelist** | Enforced β€” agent cannot use tools outside its list | | **Iteration cap** | Prevents runaway loops | | **Auto-select keywords** | Fires automatically when matched in chat | | **agentType** | Determines exhaustion behaviour (`research / execution / delivery / synthesis`) | | **Handoff instructions** | Baked into every system prompt | Profiles are editable from the UI (**Settings β†’ Agent Profiles**) or by editing `server/src/agents/seedProfiles.ts`. --- ## πŸ”Œ Integrations Add any key to `.env` β€” the agent automatically gains those tools. | Service | Env Var | Tools Unlocked | |---------|---------|----------------| | GitHub | `GITHUB_TOKEN` | search repos, list issues, create issues, get files, list PRs, add comments, search code | | Notion | `NOTION_API_KEY` | search, read page, create page, query database, append blocks | | Gmail | `GMAIL_USER` + `GMAIL_APP_PASSWORD` | send, search, read, list inbox | | Google Calendar | `GOOGLE_CALENDAR_ACCESS_TOKEN` | list events, create event, check availability, delete event | | Google Drive | `GOOGLE_DRIVE_ACCESS_TOKEN` | search, get file, list folder, create file | | Twitter/X | `TWITTER_BEARER_TOKEN` | search tweets, read timeline (+ OAuth keys for posting) | | Trello | `TRELLO_API_KEY` + `TRELLO_TOKEN` | list boards/cards, create card, move card, add comment | | Spotify | `SPOTIFY_ACCESS_TOKEN` | search, play, pause, skip, queue, create playlist | | Apify | `APIFY_API_TOKEN` | residential proxy scraping for CAPTCHA-protected sites | | Image Gen | `OPENAI_API_KEY` or `STABILITY_API_KEY` | DALL-E 3 / Stability AI image generation | **Messaging platforms:** Telegram Β· Discord Β· Slack Β· WhatsApp Β· Signal --- ## 🧠 LLM Providers At least one required. Add the key to `.env`: | Provider | Env Var | Notes | |----------|---------|-------| | Anthropic | `ANTHROPIC_API_KEY` | Claude 3.5 / 4 family | | OpenAI | `OPENAI_API_KEY` | GPT-4o, GPT-4o-mini | | Google | `GOOGLE_API_KEY` | Gemini 2.0 Flash | | Groq | `GROQ_API_KEY` | Free tier, very fast (Llama, Mistral) | | DeepSeek | `DEEPSEEK_API_KEY` | Low cost, strong at coding | | Mistral | `MISTRAL_API_KEY` | European-hosted option | | Together AI | `TOGETHER_API_KEY` | Wide open-source model selection | | Ollama | `OLLAMA_HOST` | Local models, no API key needed | | Perplexity | `PERPLEXITY_API_KEY` | Search-grounded LLM with citations | Switch models any time from **Settings β†’ Models** in the UI. --- ## πŸ“ Architecture ```mermaid flowchart TD U([πŸ‘€ User]) -->|"Natural language goal"| AO[AgentOrchestrator\nChat OODA loop] AO -->|Goal detected| GE[GoalOrientedExecutor] GE --> GX[GoalExtractor\nTyped success criteria] GX --> RR[ReActReasoner\nReason β†’ Act β†’ Assess loop] RR --> EE[ExecutionEngine] EE --> MCP[MCP Tool Servers\n17 servers] EE --> DS[Docker Sandbox\ncode execution] RR --> SC[SelfCorrector\nSemantic script repair] SC --> RR RR --> GV[GoalValidator\nChecks file / content / messaging] GV -->|All criteria pass| SL[SkillLearner\nGeneralise + store for reuse] GV -->|Criteria fail| RR style U fill:#6366f1,color:#fff style SL fill:#10b981,color:#fff style GV fill:#f59e0b,color:#000 ``` **Concurrency:** Up to 3 concurrent goals, each with up to 5 parallel sub-agents via `delegate_parallel`. **MCP Tool Servers (17):** web_search Β· browser_automation Β· code_executor Β· file_system Β· gmail Β· google_calendar Β· google_drive Β· github Β· notion Β· twitter Β· trello Β· spotify Β· apify Β· messaging (Telegram/Discord/Slack/WhatsApp/Signal) Β· image_generator Β· document_builder Β· memory --- ## πŸ—ΊοΈ Ecosystem Landscape The open-source self-hosted agent space has several strong projects, each built around a different design philosophy. Here's how they are positioned: | Project | Primary Design Focus | Best Suited For | |---------|---------------------|-----------------| | **AI Partner** | End-to-end goal execution with validated outcomes, specialist agents, document generation, and messaging delivery | Users who want to hand off a complete goal and receive a finished, delivered result β€” with no babysitting | | **Hermes Agent** | Autonomous agent with 68 built-in tools, 15+ chat platforms, NL scheduling, and RL training via Atropos | Users who want wide messaging platform coverage and self-improving skills out of the box | | **OpenHuman** | Desktop-native personal agent with 118+ OAuth integrations, auto-sync every 20 min, Memory Tree + Obsidian vault | Users who want zero server ops and automatic context from all their connected accounts | | **OpenClaw** | Personal, always-on AI assistant with 22+ messaging platforms and a modular skills registry (ClawHub) | Users who want the widest messaging reach β€” LINE, WeChat, QQ, iMessage, Matrix, and more | | **Agent Zero** | OS-level autonomy with dynamic tool creation at runtime; runs in an isolated Docker terminal | Power users who want an agent that can build its own tools and interact deeply with the operating system | | **OpenHands** | Enterprise-grade autonomous software engineering; multi-agent collaboration and audit trails | Engineering teams automating code review, bug fixing, or large-scale software development workflows | | **CrewAI** | Multi-agent crew pipelines with defined roles, tasks, and handoffs; best-documented orchestration framework | Teams building structured multi-agent workflows with clear role boundaries | > **Where AI Partner fits:** It is the only project in this space that treats goal *delivery* β€” not just task execution β€” as a first-class requirement. The agent proves outcomes via typed success criteria, validates messaging delivery, and self-corrects on failure. v2 adds human proxy capabilities β€” acting as you across email, meetings, and phone β€” which no other self-hosted tool in this table does. --- ## βš™οΈ Configuration Key files β€” editable without redeploying: | File | Purpose | |------|---------| | `server/prompts/agent.system.md` | Agent core identity | | `server/prompts/profiles/` | Per-profile LLM prompts | | `server/prompts/reasoner-reason.md` | ReAct reasoning prompt | | `server/prompts/reasoner-decide.md` | ReAct action-selection prompt | | `server/config/blocked-domains.json` | Domains blocked from browser navigation | | `server/config/data-api-hints.json` | API fallback hints injected when search fails | | `server/templates/workspace/HEARTBEAT.md` | Proactive agenda tasks | | `server/templates/workspace/SOUL.md` | Agent persona + quiet hours | **Environment variables** β€” see [`.env.example`](.env.example) for the full annotated list. --- ## 🐳 Docker Commands ```bash # Start docker compose up -d # View logs docker compose logs -f app # Stop docker compose down # Update to latest ./setup.sh --update # Mac/Linux .\install.ps1 -Update # Windows # Wipe all data and start fresh ./setup.sh --reset .\install.ps1 -Reset ``` --- ## πŸ› οΈ Development ```bash # Hot-reload dev mode docker compose -f docker-compose.dev.yml up # Run unit tests (145 tests) cd server && npm run test:unit # TypeScript check cd server && npx tsc --noEmit ``` --- ## πŸš€ What's Coming in v2 v2 shifts the platform from **working for you** to **working as you** β€” acting on your behalf across email, meetings, phone, and messaging with your identity and your voice, gated by a policy you control. | Capability | Status | |---|---| | **Human proxy** β€” reply to email, Slack DMs, Telegram DMs in your voice (not a bot) | πŸ”¨ Building | | **Meeting proxy** β€” join Teams / Zoom / Meet as a participant, speak via voice clone | πŸ”¨ Building | | **Phone proxy** β€” voicemail intake, outbound calls via Twilio | πŸ”¨ Building | | **AuthorityPolicy** β€” declarative auto / draft-and-ask / block per action + contact class | πŸ”¨ Building | | **CounterpartyStore** β€” unified identity graph across email, Slack, Telegram, and meetings | πŸ”¨ Building | | **Proactive heartbeat** β€” LLM decides when to act autonomously, no prompt needed | πŸ”¨ Building | | **Voice clone** β€” speaks in your voice via ElevenLabs / MiniMax / OpenAI | πŸ”¨ Building | | **41 MCP servers** (up from 17) β€” Stripe, HubSpot, S3, Sentry, Microsoft 365 added | πŸ”¨ Building | | **18+ LLM providers** (up from 9) β€” OpenRouter, Cerebras, NVIDIA NIM, MiniMax added | πŸ”¨ Building | | **5-layer memory** β€” adds biographic facts + counterparty graph to the existing stack | πŸ”¨ Building | **[⭐ Star this repo](https://github.com/AmitkrPaiwal/AI-Partner/stargazers)** to be notified when v2 ships. Follow progress at [aipartner.amitpaliwal.com](https://aipartner.amitpaliwal.com). --- ## 🀝 Contributing Contributions are welcome! Here are some good places to start: - Browse [**good first issues**](https://github.com/AmitkrPaiwal/AI-Partner/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) β€” beginner-friendly tasks - [**Open a Discussion**](https://github.com/AmitkrPaiwal/AI-Partner/discussions) to propose features or ask questions - Submit a PR β€” all improvements are reviewed within 48 hours **Ideas for new contributors:** - New MCP tool server integrations (Linear, Jira, Confluence, Airtable) - Additional LLM provider adapters - Browser automation improvements - UI/UX enhancements - Documentation and tutorials --- ## πŸ“„ License [MIT](LICENSE) β€” free to use, modify, and distribute. --- **Built with** TypeScript Β· Express Β· React Β· Puppeteer Β· Docker Β· SQLite Β· MCP --- ⭐ **If AI Partner saves you time, please star this repo** β€” it helps others find it. [Star ⭐](https://github.com/AmitkrPaiwal/AI-Partner/stargazers) Β· [Fork 🍴](https://github.com/AmitkrPaiwal/AI-Partner/network/members) Β· [Issues πŸ›](https://github.com/AmitkrPaiwal/AI-Partner/issues) Β· [Discussions πŸ’¬](https://github.com/AmitkrPaiwal/AI-Partner/discussions) ## Source & license This open-source MCP server is cataloged on AgentStack and links to its original source β€” we do not rehost the code. - **Author:** [Amitkrpaliwal](https://github.com/Amitkrpaliwal) - **Source:** [Amitkrpaliwal/AI-Partner](https://github.com/Amitkrpaliwal/AI-Partner) - **License:** MIT - **Homepage:** https://aipartner.amitpaliwal.com/ Install and usage instructions live in the source repository linked above. ## Pricing - **Free** β€” Free ## Security capabilities Automated source analysis of v0.1.0 β€” what this tool can access: - **Network access:** yes - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** yes - **Dynamic code execution:** no *"Yes" means the capability is present in the source β€” more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** β€” security scan: flagged β€” Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/mcp-amitkrpaliwal-ai-partner - Seller: https://agentstack.voostack.com/s/amitkrpaliwal - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack β€” the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.