Introduction

minnal (Tamil for lightning) is an open-source coding agent harness for the terminal — built for raw speed, multi-session workflows, and infinite customizability.

Where other CLIs lock you into one model and one session, minnal treats agents as a fleet: spawn them in parallel, let them collaborate, give them memory that survives across sessions, and connect them to any LLM provider you can reach.

What makes minnal different

• Lightning-fast: idle at ~28 MBRAM and a sub-100 ms cold start. Single static binary, no runtime to install.
• Multi-session by default: the minnal serve process orchestrates many sessions over a single shared core.
• Swarm: agents working in the same repo automatically detect and resolve conflicts, message each other, and spawn sub-teams.
• Real memory: a semantic memory graph recalls relevant context automatically — no more burning tokens on memory tools.
• Self-Dev: minnal can rebuild itself. Tell the agent to enter self-dev mode and it will edit, build, and reload its own binary.
• Bring your own model: Anthropic, OpenAI, Gemini, Copilot, Azure, Bedrock, OpenRouter, Groq, Ollama, vLLM, LM Studio, and any OpenAI-compatible endpoint.

Who minnal is for

Engineers who already live in the terminal and want a tool that does not get in their way. If you are comparing minnal to OpenCode, Kilo CLI, Claude Code, or pi — minnal is the option that prioritizes raw performance and harness-level capabilities (memory, swarm, self-dev) over GUI polish.

Install in 30 seconds

curl -fsSL https://minnal.dev/install | bash
minnal