Helmor Local Agent Workbench
Run coding agents in parallel in isolated Git workspaces, then review, test, and ship their work.
- Source repo
- dohooo/helmor
- Stars
- ★ 1.3k
- Last updated
- 2mo ago
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 52/100 · Major gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- A developer handling several bug fixes or features who wants a separate Git worktree and coding-agent task for each branch.
- Main limitation
- The documented release platforms are macOS and Windows x64; the supplied evidence does not document Linux support.
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
Helmor is a local-first workbench for multi-agent software development, with data and workspaces kept under `~/helmor/`. It creates a separate Git worktree and branch for each task so coding agents can work in parallel without sharing a workspace. The application brings conversations, diffs, a Monaco editor, terminals, and PR actions into one window, with GitHub and GitLab flows for creating and merging PRs or MRs, fixing CI, and resolving conflicts. It supports bringing your own Claude Code, Codex, Cursor, OpenCode, or Kimi Code credentials, API keys, and custom providers. Helmor also exposes a `helmor` CLI and an MCP server over stdio; an experimental mobile companion uses a Cloudflare tunnel to start tasks from a phone browser.
A user adds a local repository through the app or with helmor repo add /path/to/repo, then creates an isolated task branch and worktree using helmor workspace new --repo myapp. Tasks are sent to a signed-in coding agent through the interface or helmor send --workspace myapp/feature-x "...", while Helmor presents the agent conversation, code changes, editor, and terminal for that workspace. helmor workspace status checks workspace state, and helmor workspace run-action runs available delivery actions such as creating a PR, merging, or fixing CI. helmor mcp starts an MCP server over stdio, and every CLI command supports --json.
- A developer handling several bug fixes or features who wants a separate Git worktree and coding-agent task for each branch.
- An engineer using Claude Code or Codex who wants to inspect generated diffs, edit code in Monaco, and use a terminal beside the conversation.
- A maintainer working through GitHub or GitLab delivery who needs to create or merge PRs/MRs, address CI failures, and resolve conflicts from a workspace.
- A command-line-oriented developer who wants to automate local task workspaces with
helmor workspace new,helmor send, andhelmor workspace status. - A developer who wants a terminal or another agent to drive the local workbench through Helmor’s stdio MCP server.
How do you install or deploy this agent?
Download Helmor from GitHub Releases; the documented builds are macOS (Apple Silicon and Intel) and Windows x64. On first launch, connect GitHub or GitLab and sign in to a coding agent; the README states that agent CLIs are bundled. To install the CLI, use Settings → Experimental → Command Line Tool. For local source development, the documented commands are bun install && bun run dev.
How do you use this agent?
Add a repository, create a workspace, and send a task: helmor repo add /path/to/repo, helmor workspace new --repo myapp --name fix-auth, and helmor send --workspace myapp/fix-auth "Add tests for the token refresh path.". Check progress with helmor workspace status myapp/fix-auth, then use helmor workspace run-action myapp/fix-auth for an available shipping action. Workspace references use repo-name/directory-name; append --json to any command when machine-readable output is needed.
What are this agent's strengths and limitations?
- Each task receives its own Git worktree and branch, which is designed to prevent parallel agents from working over one another.
- Conversation, diffs, Monaco editing, terminals, and PR/MR actions are co-located, supporting review and shipping in one workbench.
- It explicitly supports Claude Code, Codex, Cursor, OpenCode, and Kimi Code while allowing users’ own logins, API keys, and custom providers.
- The desktop workflow is complemented by a JSON-capable CLI and a stdio MCP server for terminal and agent-driven operation.
- The documented release platforms are macOS and Windows x64; the supplied evidence does not document Linux support.
- Initial setup requires connecting GitHub or GitLab and signing in to at least one coding agent; the supplied material does not detail the authentication or permission model.
- Workspaces and data are local under
~/helmor/; there is no documented evidence here of centralized deployment, shared team state, or remote workspace management. - The mobile companion is explicitly experimental and uses a Cloudflare tunnel to reach the desktop.
- Slack and GitHub context, plan mode, and agent-driven orchestration are described as upcoming, rather than current capabilities.
How does this agent compare with similar options?
Helmor is not presented as a replacement for Claude Code, Codex, Cursor, OpenCode, or Kimi Code. It is a local workflow layer that can connect those coding agents to isolated Git workspaces, review surfaces, and PR/MR actions.
Key facts side by side with the most closely related agents.
| Agent | Source review | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|
| Helmor Local Agent Workbench This agent | 52 · Major gaps | ★ 1.3k | 2mo ago | TypeScript | Codex · Claude Code |
| Cyrus | 50 · Major gaps | ★ 825 | today | TypeScript | Codex · Claude Code · Claude API |
| Loop Engineering | 69 · Some gaps | ★ 11k | today | TypeScript | Codex · Claude Code |
| Helix Private Coding Fleet | 43 · Major gaps | ★ 809 | 5d ago | Go | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README emphasizes local-first, data stored under ~/helmor/, but permission model not detailed. User confirmation exists (e.g., before creating PR), but not all external actions. Data flow transparency limited, no details on data transmission. Sensitive data handling not specified. Dependency security: trustedDependencies and CI checks exist, but no vulnerability scan evidence. External effects: PR/MR creation etc., but user confirmation not explicit. Rollback: git worktrees and branches, but no explicit rollback mechanism. Source attribution: license and copyright present, but publisher unverified. Deductions: insufficient details on permission model, data flow, sensitive data handling.
Evidence shows: README and CI config consistent, test scripts complete, but no test results provided. Dependency availability: bun.lock and CI caching, but not verified. Failure messages: no error handling documentation. Deductions: no test results and error handling details.
Evidence shows: README describes multiple use cases (parallel agents, review, merge), but capability boundaries not explicit. Trigger precision: CLI commands and MCP server, but trigger conditions not explicit. Environment fit: macOS and Windows supported, but Linux not mentioned. Deductions: capability boundaries and trigger conditions insufficient.
Evidence shows: README structure clear, install notes (download, CLI install), naming stable (helmor), examples (CLI examples), but known limitations not explicit. License Apache-2.0, version present, changelog not provided. Maintenance responsibility: contributing guide and community links, but maintainers not explicit. Deductions: known limitations and changelog missing.
Evidence shows: Output usability: CLI and GUI, output format (--json) explicit. Marginal value: parallel agents, review, merge etc., clear value. Cost-benefit: open-source free, but no performance data. Deductions: no performance or user feedback evidence.
Evidence shows: README feature claims lack implementation details, but CI config and test files exist. Cross-source corroboration: CI and tests, but no external verification. Fact-inference separation: README claims are assertions, not distinguishing facts and inferences. Deductions: claims lack traceability.
- Publisher identity unverified, proceed with caution.
- Permission model and data flow transparency insufficient, review code recommended.
- Dependency security lacks vulnerability scan evidence, self-check needed.
- Known limitations and changelog missing, may affect maintenance.
FAQ
Am I locked into one coding agent?
Where does Helmor put task work?
~/helmor/workspaces/, with one workspace per task.Can I operate it without the desktop interface?
helmor mcp provides an MCP server over stdio.