Use Together AI Models in Hermes Agent and Hermes Desktop
Use Together's curated models in Hermes while keeping its sessions, skills, memory, preferences, terminal interface, and native desktop app.
By TogetherLink
The short answer
TogetherLink creates an isolated Hermes runtime for one launch and adds a Together provider to it. Hermes still owns the interface, tools, sessions, skills, memory, and desktop experience.
Start a first Together-backed session
Install Hermes Agent
Use the upstream project's supported installer. See the official Hermes quickstart for platform-specific options.
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashInstall and configure TogetherLink
Save a Together API key once. The selected model is billed by Together AI, not by the harness vendor.
curl -fsSL https://togetherlink.vercel.app/install.sh | bashtogetherlink configureLaunch Hermes Agent
thermestogetherlink hermes desktopWhat TogetherLink changes—and what it leaves alone
| Area | Behavior |
|---|---|
| Model requests | The harness calls Together AI directly through its OpenAI-compatible API. |
| Together key | Provided only to the launched process; it is not written into the generated provider metadata. |
| Harness state | Sessions, skills, memories, preferences, and other non-credential state are linked from the normal Hermes home and remain resumable. |
| Provider setup | A temporary HERMES_HOME contains the Together provider and isolated credentials. It is removed when the launched Hermes process exits. |
| Usage reporting | TogetherLink records launch lifecycle only. Token and cost totals are not available locally for this direct harness. |
Quit Hermes Desktop before launching the Together version
An already-running Desktop process keeps the environment it started with. Quit it completely, then run togetherlink hermes desktop so the new process receives the temporary provider and selected model.
Together changes the model, not the workflow
Terminal and Desktop
Use the CLI/TUI or the native desktop app against the same curated Together catalog.
Native sessions
Start or resume Hermes conversations while keeping the normal session store available.
Skills and memory
Hermes' native skills, memories, and preferences remain visible through the temporary overlay.
Hermes tools
Together supplies model inference; Hermes continues to run its own terminal, browser, and other enabled tools.
5 curated models, with Kimi K3 as the default
To override the model, put --model before the harness name. Arguments after the harness name belong to the underlying tool.
moonshotai/Kimi-K3Default · vision · reasoning · 1M context
zai-org/GLM-5.2text · reasoning · 512K context
deepseek-ai/DeepSeek-V4-Pro-0813text · reasoning · 1M context
Qwen/Qwen3.7-Maxvision · reasoning · 1M context
deepseek-ai/DeepSeek-V4-Flash-0731text · reasoning · 1M context
Verify the launch before starting work
TogetherLink prints a launch line before the harness opens. For the default command, look for:
togetherlink ▸ Launching Hermes with Together AI.
Launching the upstream binary directly skips TogetherLink and uses that tool's normal provider configuration.
FAQ
Can Hermes Agent use Together AI models?+
Yes. TogetherLink creates a temporary Hermes home overlay with a Together provider, then launches Hermes with the selected curated model.
Does it work with Hermes Desktop?+
Yes. Run togetherlink hermes desktop. Quit any existing Hermes Desktop process first so the app starts with the temporary Together runtime.
Will it overwrite ~/.hermes?+
No. TogetherLink copies or links ordinary Hermes state into a temporary overlay, isolates credential files and provider configuration, and removes the overlay when the launched process exits.
Are my sessions, skills, and memory available?+
Yes. Ordinary non-credential Hermes state is linked into the temporary runtime, so native sessions, skills, memories, and preferences remain available and resumable.
Does TogetherLink report Hermes token cost?+
No. Hermes calls Together directly, so TogetherLink records session lifecycle but does not see the inference stream or produce local token and cost totals.