Shelbi vs T3 Code

At a glance

Details checked November 4, 2026, against Shelbi v0.13.3 and T3 Code v0.1.7. T3 Code's README still describes the project as early and warns readers to expect bugs.

Where Shelbi excels

A task queue with explicit ownership

Shelbi separates the task from the agent session doing it. Its default board shows Backlog, Todo, In Progress, Review, Done, and Canceled. Tasks carry priorities and dependencies. The orchestrator assigns ready work to free workspaces, then picks up the next task when a slot becomes available. You can change the statuses, agent roles, and transition actions in a project's workflow.

T3 Code's documented work surface is a thread list: start a conversation, follow its work, then snooze, settle, or archive it. It also tracks delegated subagents. Shelbi is a better fit when you want a queue whose triage, execution, and human acceptance stages remain visible independently of individual conversations.

Reuse named workspaces across tasks

Shelbi's workspace pool is declared in project configuration. Each named slot belongs to a machine and keeps its git worktree between assignments. Dispatch switches branches in that checkout. Build output and installed dependencies can stay there for the next task; the agent's conversation context is cleared when switching tasks.

This makes capacity explicit: decide how many slots each host should run, then let the orchestrator fill them. Machine and workspace tags can route workflow stages to slots with the required capabilities.

T3 Code can start each parallel thread in a new worktree or reuse an existing one. It also offers optional worktree cleanup policies. Choose Shelbi when stable worker identities and reusable checkouts suit your repository better than creating worktrees as threads need them.

A review handoff you can customize

In Shelbi's default flow, a worker writes its task id to .shelbi/shelbi-ready. The hub advances the task to Review and frees the development slot for more work. Human review is a distinct stage. A configured review workspace can check out the branch and serve the app for inspection before you approve or reject it.

The orchestrator's project instructions are editable at ~/.shelbi/projects/<name>/agents/orchestrator/instructions.md. You can adapt routing and handoff policy to the project. Optional Zen Mode adds autonomy under your configured policy; review workspaces retain their human gate.

T3 Code has substantial review tooling of its own. Shelbi's advantage here is making the handoff part of the task lifecycle and workspace allocation.

Keep the agent's terminal interface

Shelbi runs agents in detached PTY sessions that outlive their clients. Use the terminal dashboard, the native macOS app, or shelbi attach <workspace> to reach a workspace's session, including on an SSH host. This suits developers who want to keep using their agent CLI's own interface while adding a board and scheduler around it.

Where T3 Code excels

More ways to work from another device

T3 Code has desktop clients for macOS, Windows, and Linux, a web client, and iOS and Android apps. The installation guide also covers running agents inside WSL. Shelbi offers a native macOS app and a terminal client on macOS and Linux.

With T3 Connect, you can reach environments without configuring router forwarding. Direct network pairing, Tailscale, and SSH routes are alternatives. Both products run work on machines you control. T3 Code makes phone and browser access a much larger part of the experience. Its host still needs to stay running and reachable.

A common interface across providers

T3 Code presents Claude Code, Codex, OpenCode, Antigravity, and other supported providers through a shared chat interface. Its author describes cross-provider delegation, including a Claude Code thread spawning Codex subagents.

That is a useful alternative to interacting with each CLI separately. Shelbi supports Claude Code, Codex, and configurable runners, but keeps their terminal interfaces. Choose T3 Code when a consistent conversation surface across providers matters more to you.

More review work inside the app

T3 Code's source control interface covers GitHub, GitLab, Forgejo, Gitea, Bitbucket, Azure DevOps, and GitCafe. It can generate commit messages and PR descriptions, show turn and branch diffs, handle review comments, request reviewers, and merge changes. Its mobile app also supports PR review.

Shelbi has workflow actions to push branches, open PRs, merge, and restack dependent work. T3 Code is attractive when you want more of the code host's review experience directly in the same app as the conversation.

Scheduled and webhook-driven work

T3 Code supports scheduled tasks and webhook-triggered prompts, including filters for incoming deliveries. Its v0.1.7 release adds one-time schedules. It also exposes thread control through its CLI and an MCP server. It can participate in automated workflows as well as interactive coding.

Choose Shelbi if…

• You want a prioritized task board with dependencies and explicit human handoffs.
• You want a named pool of reusable workspaces across local and SSH hosts.
• You want to edit the scheduler's instructions and workflow policy per project.
• You prefer your agents' terminal interfaces, with a terminal dashboard or native macOS client around them.

Choose T3 Code if…

• You want Windows and Linux desktop support, a web client, or a phone app.
• You want a common chat interface for several providers and cross-provider delegation.
• You want integrated PR review across several code hosts.
• You want scheduled or webhook-triggered agent threads.

Both projects are MIT licensed. T3 Code's FAQ says the software is free and provider usage remains your expense. T3 Connect is free for up to three linked machines at the time checked; linked paid T3 Chat plans raise that limit. Direct network access remains an alternative.