Run LimaCharlie AI Sessions on the model you choose

AI Sessions in the LimaCharlie web application now run on OpenAI, Google Gemini, and OpenRouter models in addition to Claude. Connect your own credentials, pick a provider per session profile, and the session behaves the same way regardless of which model is doing the work.

Supported providers and access paths

Sessions run on Claude by default. Beyond that, you can connect any of the following with your own credentials:

Credentials are bring-your-own-key. LimaCharlie stores them encrypted, bound to your user identity, and uses them only to run your sessions. Billing goes directly to the provider. For OpenRouter, the cost you see in LimaCharlie is OpenRouter's own billed cost for each request, so the numbers match your OpenRouter account.

Everything else stays the same

Provider choice changes the model. It does not change how the session operates:

• Tools. Terminal, shell, and file tools work on every provider with the same approval flow. allowed_tools and denied_tools patterns apply identically.

• MCP servers. Profile MCP configurations work on all providers.

• Budgets. max_budget_usd is enforced everywhere. A session that reaches its budget stops accepting prompts, on any model.

• Lifecycle. Hibernation, transparent resume, and session forking behave the same on every provider.

• Cost tracking. Usage rows carry the provider and model, so you can break spend down per provider and per model.

A few capabilities remain specific to Claude: plan mode, Task subagents, extended thinking, and full bypassPermissions. When a session on another provider requests one of these, it surfaces a clear notice rather than silently ignoring the request. The same principle applies to credentials: starting a session on a provider you have not connected fails with an explicit error instead of falling back to a different model.

Scope

Multi-provider support covers user sessions, meaning the AI Terminal and user-owned chats. Organization-owned agents launched from D&R rules run on Claude by default. The CLI's ai start-session command can launch a record's agent on any supported provider with the --provider and --credential overrides, and storing the provider choice on the agent record itself is rolling out now.

Getting started

Go to User Settings → AI Terminal in the web application. Each provider has a row with its connection status and a Connect button. You can connect several providers at once and select one per session profile, either in the profile itself or as an override on a specific session.

Full details, including API endpoints for credential management and per-provider cost notes, are in the AI Providers documentation.

If you are new to running AI operators against LimaCharlie, the Claude Code setup guide covers the ten-minute integration, and the detection engineering walkthrough shows what a working session looks like in practice.