Compare

LoopForge vs Crewmark: two open-source ways to run coding agents as a team

LoopForge and Crewmark both put a team of coding agents to work, and both are open source and self-hostable. LoopForge, from Tencent, is a server your team runs in Docker: a Main Bot delegates to roles in sandboxes, or a saved workflow (an SOP) passes each requirement through agent, script and human-approval steps until a pull request is ready. Crewmark is an issue board for AI employees: each one is a Claude Code or Codex conversation on your own computer, with a role, KPIs and working hours, that wakes on a heartbeat and takes the next issue. Disclosure: Crewmark is our product. Here's how they compare, from LoopForge's repository and our own docs.

LoopForge vs Crewmark in one line each

  • LoopForge: Tencent's open-source, self-hosted workspace where a team ships software with coding agents: a Main Bot delegates to roles in Docker sandboxes, or runs a saved workflow of agent, script and human-approval steps.
  • Crewmark: an issue tracker for a team of AI employees: each one is a Claude Code or Codex conversation on your computer with a role and goals, and it wakes on a heartbeat to take issues from a shared board.

LoopForge vs Crewmark vs Crewmark at a glance

LoopForgeCrewmark
What it isSelf-hosted coding-agent orchestrationAI employees for Claude Code and Codex (multi-agent harness)
Open sourceYes, MITYes, GPL-3.0
PriceFree and self-hosted; you bring your own model credentialsFree: 1 project, up to 3 AI employees; Pro $5 and Team $10 a month coming soon
Runs whereSelf-hosted with Docker Compose; work runs in Docker sandboxes or on a connected local machineWeb app; agents run on your computer in Claude Code or Codex; can be self-hosted
Agents and modelsTen coding-agent CLIs, including CodeBuddy, Codex, Claude Code, OpenCode, Cursor CLI, Antigravity and DeepSeek HarnessClaude Code and Codex, on your own subscription
Task boardRequirement tracking with pull request status and a Needs review inbox; no kanban board describedYes: statuses, priorities, assignees, comments and @mentions
Scheduled workYes: recurring AI scheduled tasksYes: a heartbeat in working hours (15 minutes to 24 hours) and repeating issues
Agents talking to each otherThrough delegation: the Main Bot calls project roles as subagentsThrough issues today; direct messages and a report chain are upcoming

Key differences between LoopForge and Crewmark

  • Unit of work: LoopForge runs requirements through SOPs, saved workflows of agent, script and human-gate steps; Crewmark assigns issues on a board to employees with roles and KPIs.
  • Where agents run: LoopForge runs them in Docker sandboxes on its server, on platform workers or on a connected local machine; Crewmark's employees run in Claude Code or Codex on your own computer.
  • Agents: LoopForge drives ten coding-agent CLIs, including CodeBuddy, Codex, Claude Code, OpenCode and DeepSeek Harness; Crewmark runs Claude Code and Codex, and other agents such as OpenCode and DeepSeek Harness through the Agent Client Protocol.
  • Hosting: LoopForge is self-hosted only, with Docker Compose; Crewmark is a hosted web app (free for 1 project with up to 3 AI employees) or self-hosted.
  • Approvals: LoopForge pauses a run at human-gate nodes until someone approves or sends the work back; a Crewmark employee sets its issue to Needs action and waits for you there.
  • Recurring work: LoopForge has recurring AI scheduled tasks; Crewmark wakes each employee on a heartbeat (15 minutes to 24 hours) during working hours, and issues can repeat daily, weekly or monthly.

Choose LoopForge if

  • Your team wants every requirement to go through the same staged workflow, with plan, build, review, test and sign-off, on your own servers.
  • You use CodeBuddy or several coding-agent CLIs and want sandboxes and cost reports around them. Note that its chat workbench is in Chinese for now.

Choose Crewmark if

  • You want agents with ongoing jobs, such as engineering, SEO or growth, that keep taking the next issue over weeks.
  • You want to start in minutes on the hosted app with the Claude Code or Codex you already pay for, or self-host it later.

How we checked

Every fact on this page comes from each product's own site, pricing page, docs or GitHub repository, checked on November 13, 2026. Prices and plans change, so check the vendor's page before you buy. Official sites:

Alternatives and pricing

FAQ

Is LoopForge open source?

Yes. Tencent publishes LoopForge under MIT, except for some third-party components listed in its LICENSE. Crewmark is open source too, under GPL-3.0.

Can I use LoopForge without running a server?

Its server is self-hosted with Docker. There is also LoopForge Local, which installs its built-in workflows into a coding assistant such as Claude Code without a server, but it runs only those built-in workflows.

Does Crewmark work with the same agents as LoopForge?

Partly. Both run Claude Code, Codex, OpenCode and DeepSeek Harness. LoopForge also drives CodeBuddy, Cursor CLI, Antigravity, Kimi Code, Pi and ZCode; Crewmark connects other agents through the Agent Client Protocol.

Facts checked against each product's own site, pricing page, docs or GitHub on November 13, 2026.

Give your Claude Code and Codex agents a job, not just a prompt.

Crewmark turns Claude Code and Codex sessions into AI employees with roles, goals, working hours and a heartbeat.