gptmevsHerm
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
gptme
Erik Bjäreholt · Terminal agent
#118OSSMCP
Panel
7.45 spec wins
- Reliability
- 6.7
- Usefulness
- 7.0
- Cost
- 9.0
- Longevity
- 6.8
“One environment variable gives every tool call a pleasant sound, which is one way to hear the money leaving.”
Herm
aduermael · Terminal agent
#102OSS
Panel
7.22 spec wins
- Reliability
- 7.2
- Usefulness
- 6.8
- Cost
- 8.5
- Longevity
- 6.2
“It offers an in-process Unix-like sandbox, for the days when Docker feels like too much of a commitment.”
Spec by spec
| Spec | gptme | Herm |
|---|---|---|
| Architecture | ||
| Category | Terminal agent | Terminal agent |
| Runs | local | local, sandbox |
| Platforms | macos, linux, windows | macos, linux |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | No |
| Browser control | Yes | No |
| Sandboxed execution | No | Yes |
| Multi-agent orchestration | No | YesA session can mix models by role, such as one model as the main agent, a cheaper one for exploration and another for vision. |
| Headless / CI mode | Yes | No |
| Models | ||
| Backbone | any | Anthropic, OpenAI, Gemini, Grok, OpenRouter, Ollama, Azure OpenAI, Vertex AI, AWS Bedrock |
| Bring your own model | Yes | Yes |
| Local models | Yes | Yes |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | n/a | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 4,439 | 234 |
Which one would each critic pick
| Critic | gptme | Herm | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.5 | 7.5 | no preference |
| El Crítico | 7.3 | 6.8 | gptme — The shell and Python tools run in your own environment with no container listed, which is the design and also the reason a bad command has nowhere to land but your machine. |
| El Profesor | 7.3 | 7.8 | Herm — Model assignment is per role rather than per session: a main agent, an exploration model and a vision model can each be a different provider inside one run. |
| La Inversora | 6.8 | 6.0 | gptme — One of the first agent command lines, three years of releases, one maintainer, and no company at all, which makes it durable in a way funded projects are not. |
| La Jefa | 6.8 | 6.8 | no preference |
| El Hacker | 8.8 | 8.3 | gptme — MIT, fully local through llama.cpp, plugins are ordinary Python packages, MCP servers are discovered and loaded dynamically, and there is an environment variable for tool sounds. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.