Skip to content

AgentLoom 0.1.0

Choose a tag to compare

View all tags
@impanda-cookie impanda-cookie released this 04 Aug 00:11
v0.1.0
9826f30

AgentLoom is an open-source desktop workbench that runs Claude, Codex, DeepSeek, GLM and
more — side by side, or as a team — on your own machine.

This is the first public release.

Highlights

  • Agent teams. Crown a lead (Claude or Codex) and give it a bench of cheaper models. The
    lead splits the goal, dispatches tasks in parallel, and reviews what comes back.
  • Never stuck behind one vendor. Out of quota mid-afternoon? Point the same session at a
    different provider and keep going — same thread, same context, nothing to copy-paste.
  • Built-in agent engine (myagent). A provider-agnostic agent loop written in Rust, with
    tool use, plan mode, checkpoints and event streaming. No vendor CLI required. Usable
    standalone on the command line.
  • See every move, undo any of them. Every command and file write lands as a card, backed
    by a Review panel with a file-level ledger and selective, per-file undo.
  • Session-centric workbench. All your projects in one window — GitHub repos and plain
    local folders alike — each with its own session list, groups and ⌘K search.
  • Rich transcript rendering. Mermaid diagrams, inline images, diffs, collapsible thinking
    and tool-call cards; long output folds by default.
  • Web search for every model, bring-your-own endpoints (OpenAI- and
    Anthropic-compatible, custom base URLs, local models), and English / 简体中文 in the UI.

On a 30-instance subset of SWE-bench Verified, graded by the official Docker harness with no
test leakage, myagent driving deepseek-v4-pro resolved a median of 17/30 (56.7%) across
eight runs, ranging 16–19. Full method and the exact instance IDs:
docs/benchmarks.md.

Install

Download the .dmg for your Mac below and drag it to Applications. Both builds are signed
and notarized by Apple — they open with no warnings.

  • AgentLoom_0.1.0_macOS_arm64.dmg — Apple silicon (M1 and later)
  • AgentLoom_0.1.0_macOS_x64.dmg — Intel

On Windows, build from source for now — see
Build from source. The Windows
build is experimental and has not been verified on real hardware.

Your data

AgentLoom is local-first. API keys live in your OS keychain, conversations in a database on
your own disk, and agents work directly in your own repositories. AgentLoom runs no server on
the internet; the only listeners it starts are loopback-only helpers. The only thing that
leaves your machine is the request you send to the model provider you chose. Its own
bookkeeping never touches your working tree.

Known limitations

  • The Intel build is cross-compiled and has not been verified on real Intel hardware yet.
    If you hit something on Intel, please tell us.
  • The reply-language control is a soft constraint: agents follow the interface language in
    most cases, but a long-running session can drift, and a lead's dispatched workers may not
    pick up a language you asked for mid-conversation.
  • Agent-team discussion and round-table modes are not implemented yet.
  • GitLab is not supported yet; the adapter seam exists.

Found something broken? Please open an issue.

License

AGPL-3.0. The AgentLoom and
MyAgentHubs names and logos are trademarks of MyAgentHubs and are not covered by the code
license.

Assets 4
Loading