Hecate
A personal AI workspace and runtime that acts as a control plane for agent work: a local gateway for chatting with model providers, running Hecate's own tasks, and supervising external coding-agent sessions with approvals and full tracing.
Hecate runs on a person's own machine as a place to create projects, chat with language models, supervise external agent CLIs, approve risky actions, inspect diffs and trace what happened during a run. A gateway component provides one local API compatible with OpenAI's Chat Completions and Anthropic's Messages formats, with model discovery, automatic failover between providers, rate limiting and usage visibility, alongside a React console for chats, connections, tasks, projects, usage and observability.
Its native runtime queues task runs, drives a tool-calling agent loop with per-call sandbox policy, approvals, retries and resumable execution, and can also supervise long-lived external agent CLI sessions over the Agent Client Protocol. It can run as a desktop app, from source, or in Docker, with state stored locally and its gateway bound to loopback by default; the project describes itself as pre-1.0 and not yet production-stable infrastructure.