Skip to content

forge v0.1.0

Choose a tag to compare

@hackspaces hackspaces released this 06 Jul 16:23

A model-agnostic agentic runtime for the terminal — any local model becomes a capable agent, because the intelligence lives in the harness, not the weights.

Install

pipx install blueshark-forge   # or: pip install blueshark-forge
forge setup                    # detects your hardware / picks an engine + models
forge                          # open it in any repo and talk to it

Highlights

  • Runs any local model via Ollama, or any OpenAI-compatible server (vLLM, llama.cpp, MLX, LM Studio, TGI, SGLang, cloud)
  • Harness levers that make small models capable: constrained decoding, living plan, surgical edits, self-correction, read-before-edit, a local model ladder that escalates when stuck
  • Workspace + machine awareness, streaming TUI (Esc to clear/stop), honest per-model context management
  • Native multi-agent fleet: session tracking, token-authenticated messaging, independent verification, coordination, shared learnings
  • Stdlib-only, 34 tests, CI across Python 3.10–3.13