Goose vs Hermes Agent
Goose and Hermes Agent are both free, open-source agents that run on your own machine with whatever model you connect, but they're built around different habits. Goose (Apache-2.0, now under the Linux Foundation's Agentic AI Foundation) is a desktop app, CLI and API for getting file, shell and research jobs done, with MCP extensions and shareable YAML recipes. Hermes Agent (MIT, from Nous Research) is a long-running assistant that keeps memory, writes its own skills and answers you on Telegram, Discord, Slack and other chat apps. If you plan to run a small local model, Goose is the safer bet; Hermes wants a big model.
What happened on our bench
AIDiveForge's bench ran both agents back to back on the same CPU-only machine with qwen2.5:3b, three tasks, three runs each. Goose passed 5 of 9; Hermes passed 0. Hermes sends a fixed prompt of about 11,500 tokens with every model call against about 1,150 for Goose, its Docker image was 4.17 GB against 0.71 GB, and it tried to reach 5 outside hosts to Goose's 1.
Read the task score carefully. Ollama capped context at 32,768 tokens, below the 64K Hermes' docs ask for, and a 3B model is far smaller than Hermes is meant for. So the result says which agent copes with a weak local model, not which one is better with a frontier model.
Where each one lives day to day
Goose is something you open when you have a job: the desktop app on macOS, Linux or Windows, or the CLI in a terminal. It works with 15+ providers (Anthropic, OpenAI, Google, Ollama, OpenRouter, Bedrock and others) and can use an existing Claude, ChatGPT or Gemini subscription through ACP. Recipes let you save a workflow and hand it to a teammate or run it in CI.
Hermes is more of a resident. One gateway process connects it to Telegram, Discord, Slack, WhatsApp, Signal and email, it saves memories and builds skills after hard tasks, and you choose where its commands run: local, Docker, SSH, Daytona, Singularity, Modal or Vercel Sandbox. Native Windows is supported now, not just WSL2.
Cost
Neither charges for the software. What you pay is the model bill, and Hermes' large per-call prompt means the same task costs more tokens than it does in Goose. If you're paying per token, that difference adds up over a month of use.
Goose: Pick Goose if you want an agent for file, shell and research tasks that works with a small local model or a cheap API budget, and you'd like to share repeatable recipes.
Hermes Agent: Pick Hermes Agent if you want an always-on assistant you message from your phone that remembers past work, and you'll pair it with a large model and a 64K-plus context.
Which one should you try?
Goose
FreeBest for Developers who want an open-source agent for file and shell tasks
Hermes Agent
FreeBest for People who want an assistant that remembers context across sessions
We installed and ran both
| Measured | Goose | Hermes Agent |
|---|---|---|
| Tasks passed with qwen2.5:3b (of 9) | 5 | 0 |
| Fixed prompt sent with every model call (tokens) | 1,150 | 11,500 |
| Docker image size on disk (GB) | 0.71 | 4.17 |
| Outside hosts it tried to reach | 1 | 5 |
Same model (qwen2.5:3b on 3 CPU cores), same three tasks, three runs each, graded on the files left behind. Ollama capped the context at 32,768 tokens for both, below the 64K Hermes' docs ask for, and a 3B model is far below the model size Hermes asks for, so read the task score as 'which agent copes with a small local model', not overall quality.
Test machine: 8-vCPU AMD EPYC 9354P cloud VM, each test limited to 3 cores and 10 GB RAM, no GPU, no internet (isolated Docker network). Times in seconds are from this CPU-only machine; a desktop with a GPU will be much faster. The VM is shared, so text-generation speed is shown only as a ratio between tools run back to back, not as tokens per second. Docker image size is the on-disk size from docker image ls for every tool. Local AI tools we tested · How we test · raw data
Field notes
All field notes →- Goose and Hermes: the model changed our task-test results
Goose completed five of nine attempts with our small local model and Hermes completed none. With the recorded hosted model, both completed all nine. The model and context…
- OpenCode with a 3B model: check the files, not just the exit status
OpenCode created the requested file in all three small-model attempts, but passed neither of the other tasks. Its hosted-model rerun passed all nine attempts.
Pricing, capabilities and platforms side by side. Prices are as of the "Pricing checked" date in the table.
Goose
Goose is an open-source agent with a desktop app, CLI and API, MCP extensions and reusable recipes. We ran it on three tasks with a local 3B model and with Claude Opus 5.5.
Hermes Agent
Hermes Agent from Nous Research keeps memory across sessions, saves reusable skills and can be reached through messaging apps. We ran it on three tasks with a local 3B model and with Claude Opus 5.5.
| Attribute | Goose | Hermes Agent |
|---|---|---|
| Pricing | Free | Free |
| Price | — | Free (MIT, self-hosted) |
| Pricing checked | Not checked yet | Jul 1, 2026 |
| Free trial | Not stated | Not stated |
| Open source | Yes | Yes |
| Has API | Yes | Yes |
| Self-hosted option | Yes | Yes |
| Platforms | macOS, Linux, Windows | macOS, Linux, Windows (native or WSL2), Android (Termux); runs commands locally or in Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox |
| Released | 2025 | 2026-02 |
| Pros |
|
|
| Cons |
|
|
Made up your mind?
Goose
FreeBest for Developers who want an open-source agent for file and shell tasks
Hermes Agent
FreeBest for People who want an assistant that remembers context across sessions
Specs come from each tool's own site and docs (how we test). AIDiveForge is editorially independent.