aizen
About
Aizen targets the ops-light end of the spectrum: one static binary that runs on a 512 MB VPS or Raspberry Pi with no cloud account, driven by any OpenAI-compatible endpoint (OpenAI, OpenRouter, local llama.cpp/vLLM, Anthropic gateways). The loop pairs a chat REPL with parallel reads, LSP-powered symbolic edits, and a verify gate that must pass tests and typechecks before the agent claims completion; a workflow command fans out role-scoped subagents and synthesizes the result. Sandboxing is layered under approvals — Landlock plus seccomp on Linux, Seatbelt on macOS, Job Objects on Windows — with network denied by default and API keys withheld from child processes. Extras include git-backed /timemachine checkpoints, Telegram/Discord-driven headless mode (aizen serve), and BM25 offline memory; versions after v0.5.5 are Apache-2.0, actively released (v0.6.1).