Skip to content

core-agent

core-agent

A reusable Go base agent built on the Google Agent Development Kit.

core-agent is the bottom layer for any project that needs a multi-turn LLM agent in Go. It ships the wiring — model providers, MCP servers, skills, instruction loading, permission gating, telemetry, transcript persistence — so consuming projects can focus on their own tools and product logic.

It deliberately is not a finished agent. There are no built-in bash / file / grep tools, no UI, no slash commands beyond /exit. What you get is the substrate.

Get started →   View on GitHub →

What it gives you

  • Multiple model providers — Gemini API, Vertex AI (Gemini), Anthropic / Claude (first-party + Vertex AI). Auto-detected from environment.
  • AGENTS.md instruction loading — system prompt prefix from AGENTS.md (with CLAUDE.md / GEMINI.md fallbacks).
  • MCP servers — declarative .agents/mcp.json; stdio and Streamable HTTP; namespaced and gated.
  • Claude-compatible skills — SKILL.md bundles in .agents/skills/<name>/.
  • Permission gate — ask / allow / yolo modes; built-in bash denylist; per-tool allowlists; path scope.
  • Telemetry — opt-in OpenTelemetry export (console / OTLP); off by default.
  • Headless CLI — one-shot via -p; multi-turn REPL by default.

When to use it

Use core-agent when you’re building a Go agent and you want a choice of model backends, the standard AGENTS.md / MCP / skills / permissions infrastructure without writing it yourself, and clean extension points for your own tools.

If you want a polished interactive coding agent out of the box (like cogo), this is the wrong layer — it’s the foundation those things would be built on, not a replacement for them.

Status

Early. APIs may change. Track milestones in the README’s milestone log.