Back to catalog
Orkas
MITOrkas is a local-first MIT multi-agent desktop app that coordinates 16 built-in specialists plus Claude Code, Codex, and OpenCode on macOS and Windows.
Open-source alternative to — compare it against:
Overview
Orkas coordinates a desktop team of sixteen built-in specialist agents alongside external CLI coding tools from a single workspace. Rather than locking tasks inside one chat thread, its Commander agent breaks goals into project to-dos and dispatches work in parallel across local subprocesses including Claude Code, Codex, OpenCode, OpenClaw, and Hermes-Agent.
Why Orkas instead of Claude Cowork
Hosted platforms like Claude Cowork, Manus, Cognition Devin, and Lindy.ai run tasks in cloud sandboxes, adding network latency, queue throttling, and credit markups before an agent touches your local repository or files. When a workflow spans software engineering and launch assets, cloud tools force developers to copy artifacts between web dashboards and terminals. Orkas runs on macOS, Windows, or Linux as an Electron desktop runtime where files, SQLite vector indexes, and subprocesses execute against your local disk and CPU.
From a performance standpoint, Orkas is built around context budgeting and bounded parallel execution. The Commander routes work through dispatch_to for named agents, hand_off_to for turn transfers, and run_worker for isolated subtasks, governed by session locks and concurrency limits. Long tasks compact prior steps into checkpoints while preserving file-change records, and document retrieval runs through a local 95 MB embedding model backed by sqlite-vec.
What it does well
- CLI coding agent orchestration: Drives local installations of Claude Code, Codex, OpenCode, OpenClaw, and Hermes-Agent as subprocesses so one plan delegates coding, review, and docs.
- Three-tier dispatch and compaction: Balances throughput using
dispatch_to,hand_off_to, and isolatedrun_workercalls while enforcing tool-round budgets and checkpoint compaction. - Sixteen built-in specialists: Ships roles for engineering, research, SEO, office files, video, and e-commerce that output Word, Excel, PowerPoint, PDF, and video files.
- Local vector Library and memory: Indexes PDFs and documents into
sqlite-vecwith a local embedding model for cited lookup, while separating user, project, and agent notes.
