bise is a command-line multi-agent harness that treats one “main” thread as a team lead that breaks big goals into jobs, runs specialized agents, and keeps you out of the loop until a real decision is needed. Agents work in isolated worktrees, commit their changes, and open PRs so changes flow through your usual CI and review process; they communicate with each other to avoid collisions and clean up after themselves. It supports per-agent model selection and plugins (skills, MCP servers, hooks), calls external tools by code so many integrations (GitHub, Linear, Sentry, Slack, etc.) don’t bloat context, and only spins up agents when a job requires one to limit token/billing waste. Agent messaging is provably correct via Bend, and long threads are compacted so context stays manageable even in very large repos.
Installation and use are minimal: one-line install, run in a repo (macOS supported now; Linux soon), supply git and an API key for Anthropic/OpenAI/Mistral, then talk to MAIN to set goals, check inbox, or poke individual agents. UI features include a single-thread terminal, zen typing mode, image paste and voice input, safety modes for command execution, and resume-after-restart behavior. Open source under Apache-2.0 and built by Gabriel Vergnaud, bise aims to replace session jockeying and “agent babysitting” with a lightweight, opinionated team lead that preserves developer flow.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.