ShannonvsWarren
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Shannon
Kocoro Lab · Agent harness
OSSMCP
Panel
7.03 spec wins
- Reliability
- 7.0
- Usefulness
- 6.7
- Cost
- 8.3
- Longevity
- 6.2
“It arrives as a Docker Compose stack with a workflow engine and a policy engine, so your agent framework now needs its own platform team.”
Warren
Jaymin West · Agent harness
OSS
Panel
6.82 spec wins
- Reliability
- 6.7
- Usefulness
- 7.0
- Cost
- 7.8
- Longevity
- 5.8
“There is a public instance at app.warren.run, which is a generous offer from someone who knows exactly what agents cost to run.”
Spec by spec
| Spec | Shannon | Warren |
|---|---|---|
| Architecture | ||
| Category | Agent harness | Agent harness |
| Runs | local, sandbox | local, cloud, sandbox |
| Platforms | macos, linux | macos, linux, web |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | No | Yes |
| Git operations | No | YesWarren manages git credentials, branch construction and push, and can create the pull request when configured. |
| Browser control | No | No |
| Sandboxed execution | YesCode execution is isolated in a WASI sandbox inside the Rust agent core, and the whole stack ships as Docker Compose services. | YesEach run stays inside a sandbox boundary chosen per deployment; watchdogs reconcile lost processes and pods, implying container or pod backends. |
| Multi-agent orchestration | Yes | Yes |
| Headless / CI mode | Yes | Yes |
| Models | ||
| Backbone | OpenAI, Anthropic, Google, DeepSeek, xAI, Ollama, LM Studio, vLLM | via managed agent harnesses |
| Bring your own model | Yes | Yes |
| Local models | Yes | No |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | $0/mo | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 2,279 | 468 |
Which one would each critic pick
| Critic | Shannon | Warren | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.0 | 7.5 | Warren — Pick it when an agent run needs a ceiling on what it can spend; pick Sandbox Agent if you only need the agent exposed and will supervise it yourself. |
| El Crítico | 6.5 | 6.8 | Warren — The documentation advertises watchdogs that reconcile lost processes and pods, and finalization that salvages work before teardown, which describes what happens without them. |
| El Profesor | 7.8 | 7.3 | Shannon — Running agent workflows on a durable execution engine makes every run replayable step by step, which turns debugging from archaeology into reproduction. |
| La Inversora | 6.0 | 5.8 | Shannon — A small lab, a permissive licence, 2,229 stars and no price anywhere, which means there is no business to fail and nobody obliged to keep shipping. |
| La Jefa | 7.3 | 6.3 | Shannon — Multi-tenant isolation is enforced by policy rules and destructive steps require human approval, which is the first open project this quarter that anticipated my questions. |
| El Hacker | 7.8 | 7.5 | Shannon — MIT, one install script, MCP client support, and Ollama, LM Studio or vLLM as providers, so the whole thing runs with nothing leaving the machine. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.