Planck v0.1.0

One goal, many experts. Planck is an OTP-native multi-agent coding assistant. Define a team of specialist agents, talk to the orchestrator, and let the team handle the rest.

Packages

Package	Description
planck_ai	Typed LLM provider abstraction — Anthropic, OpenAI, Gemini, Ollama, llama.cpp
planck_agent	OTP multi-agent runtime — each agent is a supervised GenServer
planck_headless	Sessions, teams, resources and config — build your own interface on top
planck	Web UI + HTTP API shipped as a self-contained binary

Install

curl -fsSL https://alexdesousa.github.io/planck/install.sh | sh

Highlights

• Multi-agent teams — static teams from TEAM.json or dynamic teams spawned at runtime by the orchestrator. Each agent is an OTP process with supervised lifecycle.
• Provider agnostic — Anthropic, OpenAI, Google, Ollama, and llama.cpp through a single API.
• Session persistence — SQLite-backed sessions survive restarts. Dynamic workers are reconstructed with their original agent IDs and full message history.
• Context compaction — LLM-based compactor summarises old messages when context grows large. Tool outputs capped at 2 000 lines / 50 KB before storage.
• Skills and sidecars — reusable prompt context via SKILL.md files; custom tools and compactors via a sidecar OTP application.
• Explicit agent targeting — ask_agent and delegate_task require identifier + identifier_type, eliminating silent targeting failures.
• Self-hosted — runs on your machine, no cloud required. API keys stay in ~/.planck/.env.