Your models.
Your way of working.
Sovereign LLM management starts with choosing where your models run. Use hosted frontier models, open-weight models through a provider, a local runtime on your hardware, or your own compatible endpoint. Partyline manages your sessions; each engine keeps its own model configuration, tools and credentials.
$ brew install partyline-sh/tap/partylineProvider subscriptions, model usage and hardware requirements are separate from Partyline.
- Hosted frontier models
Anthropic, OpenAI and Google through their engines.
- Open-weight models
Use a configured provider through opencode, goose or llm.
- Local models
Connect an engine to a compatible runtime on your own hardware.
- Your endpoint
Configure a supported endpoint in the engine, then run its sessions here.
supported engines and model routes
| Engine | Model source | Route |
|---|---|---|
| Claude Code | Anthropic | Hosted models through Claude Code |
| Codex | OpenAI | Hosted models through Codex |
| Gemini CLI | Hosted models through Gemini CLI | |
| Antigravity | The agy command-line engine | |
| opencode | Configured provider/model | Hosted, open-weight and compatible local endpoints |
| goose | Configured provider | Local or hosted models supported by goose |
| prime-agent | Prime Intellect | Models available through prime-agent |
| llm | Configured model and plugins | Compatible endpoints; Ollama, llama.cpp and MLX through plugins |
how it works
configure your engine
Install and authenticate the engine you want to use. Select its provider, model or local endpoint using that engine’s own configuration.
run sessions together
Keep sessions from different engines in the same terminal workspace. Each retains its native tools, permissions and provider connection.
take control from your phone
Keep sessions on your computers and connect through a relay you control. The mobile app reads the terminal and sends input to your paired computers.
what you're wondering
does a private relay make my model local?
No. The relay transports encrypted session and chat content. A hosted model still receives the prompts and context your engine sends it. Local inference requires a local runtime configured in that engine.
is every capability identical across engines?
No. Resume, streaming, image review, tool integration and other capabilities vary. Working and needs-you detection currently reads Claude Code and Codex screens; other engines can appear idle while running.
which individual models are supported?
The model catalogue belongs to your engine and provider. Partyline accepts the model choices supported by its adapters; access, runtime compatibility and tool support depend on the model and engine you configure. It does not ship model weights or an inference server.
Choose the model. Keep the workspace.
$ brew install partyline-sh/tap/partylineUse shared context across sessions, then connect through your own relay.