Lody
Share, dispatch, and supervise your team's existing coding agents across desktop, mobile, web, and CLI.
- Source repo
- LodyAI/Lody
- Stars
- ★ 1.1k
- Last updated
- 1d ago
- License
- Apache-2.0
- Primary language
- TypeScript
- FA score
- 75/100 · Good
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A development team already using Codex, Claude Code, Kimi, or OpenCode can share live conversations instead of circulating screenshots and logs.
- Not a fit if
- Teams that require a fully local-first architecture today
- Environments that must operate without networking or sign-in
- Users seeking a bundled coding model rather than bringing an existing Agent
- Source review
- 75/100 · Good
What does this agent do, and when should you use it?
Lody is a shared workspace for connecting a team's existing coding agents and the machines that run them; it is not presented as a new coding model. Through ACP, it works with Claude Code, Codex, Kimi, OpenCode, and other compatible agents while retaining each machine's existing subscriptions, logins, models, and permission modes. Users can create and follow sessions from desktop, mobile, web, or CLI, with access to transcripts, runtime status, files, code changes, and permission requests. A machine joins the workspace through `npx lody daemon start`, while the CLI can register projects, inspect machines and Agent configurations, create sessions, and archive or restore them. Parallel jobs can use separate Git worktrees, and an Agent can coordinate investigation, implementation, and testing through independent child conversations. The project uses CRDT-based synchronization and is working toward full local-first support, but the README says that transition is not yet complete.
Lody runs lody daemon on a workstation, server, or cloud VM and connects that machine to a workspace through a sign-in link. It reads and exposes registered project files, Git diffs, session history and status, linked repositories, machine details, and Agent configurations; its review interface also places pull request, CI, and GitHub review information beside the conversation. Users can invoke lody session create with a workspace, Agent configuration, and repository, inspect sessions with lody session list, and message, cancel, archive, restore, or fork sessions. A coordinating Agent can create or reuse other conversations, read their status and history, send follow-up instructions, and bring results back, while separate Git worktrees isolate concurrent edits. Desktop, mobile, web, and CLI surfaces share progress, permission requests, attachments, and code changes; the built-in web preview can also send element-level visual annotations to the Agent.
- A development team already using Codex, Claude Code, Kimi, or OpenCode can share live conversations instead of circulating screenshots and logs.
- A team with workstations, servers, or cloud VMs can dispatch coding work remotely and approve permission requests from a phone or desktop.
- An engineering lead investigating a defect can ask a coordinating Agent to separate analysis, implementation, and testing into parallel conversations.
- Developers exploring multiple approaches in one repository can isolate each session in its own Git worktree.
- Reviewers can examine files, diffs, pull requests, CI status, and GitHub threads beside the Agent conversation that produced the change.
- Frontend teams can open a running application, switch responsive viewports, and return element-level visual annotations to the Agent.
How do you install or deploy this agent?
The source does not provide complete commands for building every application from the repository. Its documented first connection path requires a terminal capable of running npx and completion of the sign-in flow opened by the command. Start the machine daemon with:
npx lody daemon startThis opens a sign-in link, connects the workstation, server, or cloud VM to the workspace, and leaves it available for work dispatched from desktop, mobile, web, or CLI. The required Node.js version, account-creation procedure, and self-hosted backend deployment steps are not specified.
How do you use this agent?
After connecting and signing in on the machine, make sure the intended project, workspace, and Agent configuration are available. This example creates a task using the codex configuration in the my-team workspace against owner/repo:
npx lody session create --workspace my-team --agent-config codex \
--repo owner/repo "Fix the failing test"List sessions in that workspace with:
npx lody session list --workspace my-teamOpen the same session from desktop, mobile, web, or CLI to inspect its state, history, files, and diffs or add instructions. For parallel work, assign separate Git worktrees or let a coordinating Agent create, reuse, and monitor other conversations. CLI commands that support --json can supply workspace data to scripts and other tools.
What are this agent's strengths and limitations?
- ACP support connects existing Claude Code, Codex, Kimi, and OpenCode setups while preserving their subscriptions, logins, models, and permission modes.
- One session remains visible across desktop, mobile, web, and CLI, including conversations, permission requests, progress, and code changes.
- Agents can create, reuse, inspect, and direct other conversations, enabling parallel investigation, implementation, and testing.
- Separate Git worktrees, adjacent diffs, line comments, pull request state, and CI visibility keep execution and review in one context.
- Mobile notifications, diff inspection, permission approval, and iOS Live Activities support supervision away from the development machine.
- The project explicitly says it is still moving toward full local-first support, so adopters should not assume that architecture is complete.
- A machine must be connected through a sign-in link; no offline or account-free operating path is documented.
- Lody expects users to bring and configure an ACP-compatible coding Agent rather than supplying a bundled model or standalone coding engine.
- The source does not specify the Node.js version, backend self-hosting procedure, authentication details, or service pricing.
- Stable releases do not build or upload installers or auto-update artifacts; the documented release workflow only synchronizes versions and creates a GitHub Release.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Lody This agent | 75 · Good | Self-hosted service | ★ 1.1k | 1d ago | TypeScript | Codex · Claude Code |
| Alas Agent Workspace | 78 · Good | Desktop appFree + model costs | ★ 18 | today | Swift | Codex · Claude Code |
| Happier | 82 · Good | Desktop appFree + model costs | ★ 1.7k | today | TypeScript | Codex · Claude Code |
| Ateam Agent Workbench | 70 · Some gaps | Desktop appFree + model costs | ★ 12 | today | TypeScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
CI explicitly uses read-only contents permission and disables credential persistence. The README says machines remain private until shared, supports configured permission modes, and exposes permission requests, transcripts, status, files, and diffs across clients. Archive/restore, Git worktrees, and session forks provide partial isolation and recovery. Deductions apply because destinations, retention, encryption, and deletion behavior for cloud/CRDT synchronization, attachments, transcripts, repository content, and credentials are not specified; SECURITY.md governs vulnerability reports rather than providing a complete runtime-data policy. Dependency installation uses a pinned pnpm release, integrity digest, and frozen lockfile, but no audit, SBOM, or vulnerability-remediation evidence is supplied. The author label, project site, repository identity, and Apache-2.0 terms provide attribution, though the publisher is unverified and the license appendix does not identify a concrete copyright holder.
The README, package scripts, CI workflows, and ACP tests consistently describe the product structure, model capabilities, session history, and steering protocol. Tests also cover malformed metadata and fail-closed configuration mismatch behavior, supporting full self-consistency. Dependencies have version constraints, a locked installation path, and an explicit Node range, but the product relies on external agents, ACP, app stores, and optional cloud services without documented degradation strategies. CI preserves failure artifacts and reconciles regression issues, while evidence for end-user runtime error messages, retries, and offline failure handling remains limited.
The sources clearly address team sharing, remote dispatch, parallel development, review, and visual-feedback scenarios across desktop, mobile, web, CLI, multiple machines, and several ACP agents. Agent Roles, model/permission/instruction presets, and capability-normalization tests show meaningful configurability. Deductions apply because compatibility remains constrained by ACP and specific adapters, with no complete support matrix, unavailable-capability behavior, or comprehensive permission boundaries. Dispatch and steering have explicit session/configuration semantics, but automatic triggers, scheduling conflicts, and false-trigger prevention are not comprehensively documented.
The README clearly organizes capabilities, connection, CLI use, coordination, repository layout, and releases, with stable workspace/session/ACP terminology. It provides an npx quick start and concrete CLI examples plus links to fuller documentation and CONTRIBUTING, but the supplied evidence lacks complete development-install instructions and an FAQ. It candidly states that full local-first operation is unfinished and documents release/CI constraints, though there is no systematic limitations catalog. The complete Apache-2.0 text earns full license credit. Stable tags and generated release notes establish a versioning process, but no actual changelog is included. Security reporting, community channels, and maintainer actions are identified, while named maintainers and support commitments are absent.
Complete conversations, runtime state, files, diffs, permission requests, terminal access, previews, and cross-device controls make agent output directly usable in collaboration and code review. Multi-session coordination, worktree isolation, and reuse of existing agents add clear value over exchanging screenshots or logs. Cost-benefit is deducted because the source only mentions reusing subscriptions and viewing token, quota, and resource usage; it does not quantify deployment, cloud synchronization, always-online machine, mobile, maintenance, or scale costs.
Core structure and selected behaviors are traceable across the README, package manifest, CI configuration, and tests for history merging, capability normalization, and steering. Roadmap language is generally separated from current behavior with terms such as “plan,” “goal,” and “moving toward.” Deductions apply because many product-experience, security, and cross-device synchronization claims remain README assertions; the supplied code does not cover authentication, storage, network transfer, permission execution, or recovery implementation, and no independent source or release artifact corroborates them.
- Before connecting private repositories or real credentials, verify transfer destinations, encryption, retention, deletion, and administrator-access rules for conversations, attachments, source code, tokens, and CRDT/optional-cloud data.
- Confirm how mobile or remote approvals bind to the underlying agent permission mode, and test that denial, timeout, disconnection, cancellation, and concurrent steering actually fail closed.
- Review the complete lockfile, submodules, packaged artifacts, and dependency vulnerability state; the supplied evidence establishes locked installation and CI checks, not an absence of supply-chain risk.
- Validate the recovery boundaries of worktrees, archive/restore, and cancellation on critical repositories; these features do not amount to transactional rollback of arbitrary external agent effects.