Open Source Radar — October 18: Orca, AI memory, agent-native CLIs
Today's daily trending board belongs to the layer around the model: a fleet manager with nearly 100,000 stars, a memory server built for switching vendors, and a registry that gives GUI-only software an agent interface. Not a single new model among them — the movement is in the plumbing.
Orca (TypeScript, 95,844 stars) — an open-source desktop app for running a fleet of coding agents side by side: Codex, Claude Code, OpenCode or any terminal-bound agent, each in its own git worktree so five parallel attempts at one task never stomp each other, then you compare the diffs and keep the winner. It sits at 95,844 stars on today's daily board, created in March 2026 and shipping its 973rd release — the cadence of a product rather than a weekend project. The parts you would actually use: a phone companion app for steering agents away from the desk, worktrees on a remote box over SSH, and comments on AI diffs that feed straight back to the agent. Reach for it when one agent at a time has become the bottleneck.
ai-memory (Rust, 10,145 stars) — long-term memory that survives switching vendors: quit Claude Code mid-task, start Codex in the same directory, and the next agent inherits where you left off — what failed, what is still open — because twenty-plus harnesses feed one shared server. The design bet is the inverse of the usual vector-store stack: the source of truth is a git-backed wiki of plain markdown you can grep and hand-edit, the index is rebuildable from those files, and capture, search and handoffs all run with no LLM calls by default. Handoffs are typed and claimed exactly once, so two sessions cannot both grab the baton. Use it if you bounce between coding agents and are tired of re-explaining your architecture every Monday — the same itch Claude-Mem scratches, but built for cross-vendor and team use.
CLI-Anything (Python, 54,280 stars) — a HKUDS project built on a blunt premise: today's software serves humans, tomorrow's users are agents, so every desktop app needs a command line an agent can drive. It ships tested harnesses for programs that never had one — Blender, Calibre, Rekordbox and draw.io among them, each packaged with the skill documentation an agent needs to operate it correctly, and a test suite the README reports at 2,461 passing checks. There is even a demo where an agent builds an orbital drone scene in Blender end to end, and a technical report on arXiv documenting the approach. The stars are climbing faster than the commits right now (last code push was September 22), which is what a wave of new users adopting a stable tool looks like — reach for it when an agent needs software that only has a GUI.
Worth watching this week: fleets of agents are multiplying fast, and none of them share a brain yet — the memory layer is the piece nobody has standardized on.
If three agents can work on one repo at once, whose memory should they keep? Write to us via our contact page. Sources: Orca (GitHub) · CLI-Anything (GitHub) · ai-memory (GitHub)
