In Development

AI agent runner.
Spec-driven. Verified.

Run coding agents with any model, any provider. Merlin reads your specs, writes the code, and verifies the result before reporting done.

See What's Different Read the Docs

merlin

# Merlin reads your specs before writing code

> merlin "Add rate limiting to the API"

Reading specs/api/rate-limiting.md...

Found: 100 req/min per key, sliding window, Redis-backed

Planning implementation across 3 files...

Created src/middleware/rate_limit.rs

Updated src/routes/mod.rs

Added tests/rate_limit_test.rs (4 tests passing)

All changes match spec requirements. Ready to commit.

31 LLM ProvidersSpec-DrivenFledge ProtocolMulti-Channel

Spec-Aware Planning

Merlin reads your specs before writing code. Constraints go into the system prompt as hard rules.

Any Model, One Interface

OpenAI, Anthropic, OpenRouter, Groq, Together, and Ollama Cloud (Qwen3-coder, Kimi, GLM, MiniMax, GPT-OSS, DeepSeek, Devstral, Gemma). Switch providers with one flag.

Sub-Agents Keep Context Small

subagent-spawn delegates self-contained subtasks to child Merlin processes. The parent sees a compact summary, not the child's full transcript — so a fanout over 50 items keeps the parent's working memory roughly flat.

Plugin Architecture

Bundled plugins for files, search, shell, git, vision, voice, sub-agent delegation, and the Discord + Telegram bridges. Write your own in any language.

AI agent runner.Spec-driven. Verified.

Spec-Aware Planning

Any Model, One Interface

Sub-Agents Keep Context Small

Plugin Architecture

Ready to build spec-first?

AI agent runner.
Spec-driven. Verified.