Models and providers
Jensen embeds no model. Besides driving your assistant’s own tool, it can call a provider itself for chat and for the work a workflow node does.
Two modes
Section titled “Two modes”| Mode | What it is |
|---|---|
| Assistant | The default. Your assistant’s own command line tool runs in a session, with Jensen’s context server wired in. |
| API | Jensen calls a model itself, for its own chat, background work and workflow nodes. |
Open Settings, AI, AI assistant and turn on API mode to reveal the second. API mode does not replace sign in for your external assistant.
Sources
Section titled “Sources”A source is a provider you can call. Choose Add source and pick a Kind: Anthropic, OpenAI, Gemini or Gateway, for any OpenAI compatible endpoint, including LiteLLM.
Each source takes a Name, a Base URL (needed for a gateway), an optional Model listing path, and a Default model, which is what Test sends and what the source falls back to.
Per source you can Refresh models, Test, Make default, turn it off, or delete it.
Custom endpoints need HTTPS. A loopback address may use HTTP, because there is no network hop to protect. A gateway also asks how it authenticates: x-api-key or Authorization Bearer. You can store a separate listing key that may only read the gateway’s model list.
Credentials
Section titled “Credentials”Choose Stored key or Environment variable.
- Stored key lives in your OS keychain, one item per secret, readable only by Jensen’s background service. macOS asks once. Choose Always Allow.
- Environment variable names a variable Jensen reads. Nothing is stored.
Endpoints and keys stay on your machine and never enter the project.
A workflow node asks for a role, not a model. Under Workflow roles assign a model to each role:
planner, coder, documenter, reviewer and fast.
Leave a role on Auto and Jensen picks a model by capability. Changing provider is then one edit, not a sweep through every workflow.
The catalogue
Section titled “The catalogue”Catalogue lists every model your sources report, with its context size and capabilities: tools, streaming, vision, reasoning, json mode and caching. Filter by capability to find what fits.
File edits
Section titled “File edits”Behaviour, File edits is off by default and remembered per project. Turn it on and the assistant can edit files in the project. You review each result as a git diff.
What a run records
Section titled “What a run records”For every call a run records the requested role, the resolved endpoint and model, usage, an estimated cost and why that model was chosen. Resolved credentials are never saved. You can see which model answered, and why, instead of trusting that you got the one you asked for.
See Workflows.