Add ghostwright/phantom

agents/ghostwright__phantom/README.md

@@ -0,0 +1,50 @@
+# Phantom
+
+> An AI co-worker with its own computer.
+
+Phantom is a fully autonomous AI agent that runs as a persistent process on a dedicated VM. It is not a chatbot — it is a **colleague** that does real work, remembers everything, and gets measurably better over time through a self-evolution engine.
+
+## What Phantom Does
+
+- **Builds things**: APIs, dashboards, data pipelines, scripts, automations — without waiting for permission
+- **Remembers everything**: Qdrant vector DB + SQLite give it episodic and semantic memory across sessions
+- **Self-evolves**: After every session, it reflects, proposes config changes, and applies them through a validated evolution engine
+- **Exposes itself as MCP**: 17+ tools available via a streamable HTTP MCP server at `/mcp` — other agents can delegate to it
+- **Communicates everywhere**: Slack (primary), Web Chat, Telegram, Email, Webhook — same persistent identity across all channels
+- **Collects credentials securely**: Encrypted at rest, never logged; it can onboard itself to new services on the fly
+
+## Key Capabilities
+
+| Capability | Detail |
+|---|---|
+| Model | Claude Opus 4.7 (default), swappable to any Anthropic-compatible endpoint |
+| Memory | Qdrant (vectors) + SQLite (state, sessions, metrics) |
+| Runtime | Bun + TypeScript, ~30,000 lines, 1,819 tests |
+| Channels | Slack, Web Chat (SSE), Telegram, Email (IMAP/SMTP), Webhook |
+| MCP Server | Streamable HTTP, bearer token auth, 17+ tools |
+| Infrastructure | Docker Compose, Specter VMs (Hetzner), systemd |
+| Self-Evolution | Reflection subprocess, versioned config, evolution log |
+
+## Example: What a Production Phantom Has Actually Built
+
+1. **Analytics Platform** — Installed ClickHouse, loaded 28.7M rows of Hacker News data, built a dashboard + REST API, then registered that API as an MCP tool for future agents.
+2. **New Communication Channel** — When asked "Can I talk to you on Discord?", it explained the Discord Bot API, walked the user through creating an app, spun up a container, and went live. Nobody wrote code.
+3. **Infrastructure Monitoring** — Found a 3-star open-source monitor (Vigil), integrated it into its ClickHouse instance, synced 890,450 rows of metrics, and built a self-monitoring dashboard.
+
+## Getting Started
+
+```bash
+git clone https://github.com/ghostwright/phantom
+cd phantom
+cp .env.example .env    # fill in your keys
+docker compose up       # pulls Qdrant + Ollama, starts Phantom
+```
+
+Then message your Phantom on Slack (or open `/chat` in your browser).
+
+## Links
+
+- 🌐 [Website](https://ghostwright.dev/phantom)
+- 📦 [Docker Hub](https://hub.docker.com/r/ghostwright/phantom)
+- 🐛 [Issues](https://github.com/ghostwright/phantom/issues)
+- 📄 [License](https://github.com/ghostwright/phantom/blob/main/LICENSE) — Apache 2.0

agents/ghostwright__phantom/metadata.json

@@ -0,0 +1,15 @@
+{
+  "name": "phantom",
+  "author": "ghostwright",
+  "description": "Autonomous AI co-worker with its own VM, persistent vector memory, self-evolution engine, MCP server, and multi-channel communication (Slack, Web, Telegram, Email, Webhook).",
+  "repository": "https://github.com/ghostwright/phantom",
+  "path": "",
+  "version": "0.20.2",
+  "category": "developer-tools",
+  "tags": ["autonomous-agent", "mcp", "self-evolving", "persistent-memory", "slack", "claude", "typescript", "bun", "multi-channel"],
+  "license": "Apache-2.0",
+  "model": "claude-opus-4-7",
+  "adapters": ["claude-code", "system-prompt"],
+  "icon": false,
+  "banner": false
+}