AI

Relay assistant

Relay is Synapse's built-in AI chat: bring your own API key, ask it about a session, and let it type commands for your approval.

Relay is a panel docked to the right of the window, available for every session - it isn't a service you need to switch on. Open or close it with Show or hide Relay (More menu, command palette or its shortcut, which has no default key until you set one in Settings > Keyboard shortcuts). Drag its left edge to resize it, or expand it to the full window width.

Adding a provider

Open Relay settings (the gear icon in Relay's header) to add an API key. Synapse supports:

• Anthropic, Google Gemini, OpenAI and xAI Grok, each with its own model picker
• Ollama, for a model running on your own machine or network - no API key needed
• A custom OpenAI-compatible endpoint, for services like Groq, OpenRouter, vLLM or LM Studio

API keys are stored in the vault; Relay settings shows whether they're encrypted (a master password is set) or not. If you pick a model with no key yet, Relay shows a form for it right there in the chat, with a link to get one from the provider.

There's also a Built-in playbooks option that isn't AI at all: it recognises a few keywords (health, disk, ports, docker, logs) and proposes fixed diagnostic commands without sending anything to a provider.

Attaching context

Relay doesn't see your terminals by itself - you choose what to share, from the composer's "Add context" menu:

• Active terminal output - the end of the focused session's scrollback
• Saved host list - names and addresses only, never credentials

Other features can attach things to Relay for you too, such as a screenshot from Web preview.

Running commands

By default Relay works with one session at a time, shown as a chip above the composer - pick which session a reply should run in. Every command Relay proposes appears as a card you review before it runs.

A mode switch lets you choose Confirm (nothing runs until you press Run on a step - the default and the recommended setting) or Auto-run (Relay types every proposed command straight into the real terminal without asking). Auto-run resets to Confirm every time Synapse restarts, and a warning banner shows above the composer whenever it's on.

Relay across sessions

With the Agent workflows service on, a setting called Relay across sessions lets one Relay conversation see every open session at once and propose a step for any of them - each step still names its session and still needs your confirmation. Without it, Relay stays scoped to the one session you've picked.

Guardrails and usage

Two more settings live inside Relay settings once their services are on:

• AI guardrails (on by default) hides secrets from anything sent to a provider and always confirms risky commands.
• AI usage and limits tracks token use and cost per day, and shows a countdown when a provider's usage limit is hit.

See Services for what each one covers.