EloPhanto – AI agent that runs locally
Ambitious local agent that builds own tools, but promises outpace implementation transparency.
Open-source autonomous AI agent with a self-model that actually changes as it runs - ego, affect, and identity grounded in real psychology (Higgins, PAD/OCC). Ships code, runs a team, makes money. Local-first, self-custody.
Agent that writes its own tools end-to-end when stuck. 99+ self-built tools prove the concept works.
Developers wanting local AI automation with full code visibility; power users seeking autonomous tool-building agents
Claude Computer Use · Anthropic Codebase Q&A · Cursor IDE (with agent mode)
The standout feature: when EloPhanto encounters a task it doesn't have a tool for, it autonomously writes the Python code, tests it, reviews itself, and integrates the new tool permanently. It's now built 99+ tools for itself this way.
Other features: - Multi-channel gateway (CLI, Telegram, Discord, Slack) with unified sessions - MCP tool server support (connect any MCP server) - Document & media analysis (PDF, images, OCR, RAG) - Agent email (own inbox for service signup/verification) - Crypto payments wallet (Base chain, spending limits) - TOTP authenticator (autonomous 2FA handling) - Evolving identity that learns from experience - Skill system with EloPhantoHub marketplace (28 bundled skills)
It's open source (Apache 2.0), local-first, and designed to be your personal AI operating system.
The project is very new — currently at 6 stars on GitHub. I'd love to get feedback on the architecture, the self-development approach, or what features you'd want in a local agent.
Ambitious local agent that builds own tools, but promises outpace implementation transparency.
Turns pass/fail eval signals into reusable skills without retraining the model.
Self-writing book on agentic patterns that dogfoods its own subject matter.
Sandboxed agent that writes its own Python tools and remembers mistakes in JSON.
Self-healing Playwright tests via AI agent repair, but Playwright Inspector and Cypress already auto-fix selectors.
Autonomous agents evolve batch correction algorithms to super-human performance without human intervention.