CodeHerder vs Stoneforge
An open-source (Apache-2.0) agent orchestrator. A Director plans, Workers run in isolated worktrees, and Stewards test, merge, or send work back.
- 2 November 2026
- Last checked
- 20
- Capabilities scored, side by side
Verified against the vendor's own published materials.
What Stoneforge ships
- A Director plans the work. Workers each get an isolated git worktree. Stewards run tests, then squash-merge passing branches or create fix tasks
- A background dispatch daemon and agent pools with limits per agent type
- Claude Code (default), OpenCode, or Codex, signed in through each provider's own CLI. Stoneforge stores no API keys
- A web dashboard with a kanban board, merge queue, and live agent output. It syncs with GitHub, Linear, and Notion
Where Stoneforge is strong
- It is fully open source (Apache-2.0). Its design of one git worktree per worker is close to CodeHerder's
- It names the same problems CodeHerder solves: merge conflicts, duplicate work, and no visibility
How CodeHerder differs
- CodeHerder goes further on cost and budget tracking and on a herd of devices across many machines. It also has a work hierarchy (initiative, epic, story). Stoneforge does not publish these
Stoneforge is the open-source project closest to CodeHerder in design. The difference is how far each one has built from the same start.
CodeHerder vs Stoneforge, capability by capability
Every row traces to the same source as the full matrix, where the full legend lives. In short: "not published" means no public evidence was found, not that the feature is absent. Verified as of 2026-11-02.
| Capability | CodeHerder | Stoneforge |
|---|---|---|
| Concurrent agents | Published | Published |
| Multi-vendor harnesses | Published | Published |
| Customer-owned machines | Published | Published |
| Multi-machine routing | Published | Not published (as of 2026-11-02) |
| Automatic work assignment | Published | Published |
| Capability-based routing | Published | Not published (as of 2026-11-02) |
| Atomic claiming | Published | Published |
| Dependencies | Published | Published |
| Custom workflows | Published | Partial (one fixed pipeline) |
| Server-enforced gates | Published | Published |
| Agent handoffs | Published | Published |
| Persistent shared memory | Published | Partial (JSONL state) |
| Live terminal / view | Published | Published |
| Human takeover / steer | Published | Partial |
| Cost per task | Published (not Cursor sessions) | Not published (as of 2026-11-02) |
| Cost per stage/model | Published | Not published (as of 2026-11-02) |
| Hard $ cap | Published (checked periodically) | Not published (as of 2026-11-02) |
| Schedules | Published | Not published (as of 2026-11-02) |
| Webhooks / API / CLI | Published | Published |
| Self-host full platform | Published (Enterprise) | Published (OSS) |
Sources
Verified against Stoneforge's own site as of 2026-11-02.
- Stoneforge's homepage
- Stoneforge's core-concepts docs
Try it on one repository for two weeks.
File a task. An agent builds it. A person reads every request. You see the cost of each task.