This is an early release preview. You may encounter bugs.
Agenvoy logo
Unclaimed

Tool coding

Agenvoy

Make AI actually work for you - A personal agent that writes its own tools and repairs itself.

A 80/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About Agenvoy.

Agenvoy is a self-hosted personal agent that runs as a daemon on your own computer and ships as a single Go binary. It breaks a request into steps, calls tools, and reports the result, and it keeps sessions, memory, schedules, and tools on local disk in SQLite with API keys and OAuth tokens in the OS keychain. A terminal interface and a web dashboard served by the binary show command output as it streams.

When no suitable tool exists, the agent writes one as a JavaScript or Python script, tests it in a sandbox, and keeps it for reuse, and it repairs tools after failed runs. Commands run under sandbox-exec on macOS or bubblewrap on Linux with network access off unless approved for that call, and sensitive paths need confirmation.

Agenvoy is also an MCP server: Claude Code, Codex, OpenCode, and other MCP clients connect to it and share the same sandboxed tool library, including tools one agent built for another. It routes tasks to models by tier across providers, signs in with existing subscriptions such as GitHub Copilot or OpenAI Codex, and can point at Ollama or LM Studio for local models. Telegram and Discord bots reach the daemon through outbound connections, so the host does not open inbound ports.

  • Tool generation: builds, sandbox-tests, and saves a new tool when none exists, and fixes tools after execution failures
  • Shared MCP tool library: exposes sandboxed tools as an MCP server to Claude Code, Codex, OpenCode, and other agents
  • Scheduling: creates recurring jobs from one sentence and pushes results to a chosen channel
  • Long-term memory: summary-based memory that retains key information across sessions
  • File search: answers questions from local files
  • Sub-agents: splits multi-step work across sub-agents and tracks their action logs
  • Model routing: routes each task to a model tier across providers, with configurable fallback priority
  • Sandboxed execution: sandbox-exec on macOS and bubblewrap on Linux, with network off by default and confirmation for sensitive paths
  • MCP client: connects to stdio and HTTP MCP servers with OAuth login and live tool refresh
  • Chat channels: Telegram and Discord bots over outbound connections only
  • Web dashboard and voice: manages sessions, tools, schedules, and memory from a browser, with interruptible voice conversation
  • Tool market: shares and installs tools, plus image generation, transcription, and text-to-speech through configured providers

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with Agenvoy.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Agenvoy →

Browse the catalog