Skip to content

Set up AI

Oleafly’s AI is strictly bring-your-own: you connect a provider you already have (or a local model), your key stays on your machine, and requests go directly from the app to that provider. There is no Oleafly server in the loop and no markup on your tokens.

Everything here lives in Settings, AI Assistant, also reachable from the chat panel’s Connect a provider button.

Provider Models offered
OpenAI GPT-4o, GPT-4o mini, GPT-4.1, GPT-4.1 mini, o3-mini
Anthropic Claude Sonnet 4, Claude 3.5 Sonnet, Claude 3.5 Haiku
Z.AI (GLM Coding Plan) GLM-5.2, GLM-4.6, GLM-4.5 Air, GLM-4.5
Groq Llama 3.3 70B, Llama 3.1 8B Instant
OpenRouter GPT-4o mini, Claude 3.5 Sonnet, Gemini Flash 1.5, Llama 3.3 70B
DeepSeek DeepSeek V3 (chat), DeepSeek R1 (reasoner)
Mistral Mistral Large, Codestral, Mistral Small
xAI (Grok) Grok 2, Grok Beta
Ollama (local) Whatever you’ve pulled: Llama 3.2, Qwen 2.5, Mistral, Gemma 2, …

Each provider card carries a Get key link straight to the right page. Reasoning models (GLM, DeepSeek R1) stream their thinking phase live into the chat; see Chat & tools.

  1. Expand the provider’s card and paste your API key.
  2. Save. The provider becomes Active immediately (green badge) and a Model dropdown appears; pick the model you want as default.
  3. Done. The chat panel header now shows your model.

Configure as many providers as you like: extras show a grey Connected badge with an Activate button, and the chat panel’s model menu switches between all of them on the fly. The trash icon deletes a key.

Provider cards in Settings, AI Assistant

Zero-cloud AI in three steps:

  1. Install Ollama and pull a model: ollama pull llama3.2
  2. In the Ollama card, click Check for Ollama. Oleafly detects the local server and lists your installed models.
  3. Pick one. That’s the whole connection; no key exists because nothing leaves your machine.

The card handles the failure case too: if Ollama isn’t running you get the exact commands to fix it and a Re-check button. A custom host URL (for Ollama on another machine) hides under “Change host (advanced)”.

The Custom instructions box appends your standing preferences to every AI request, chat and inline edits alike: “Always write in British English. Keep explanations short. Prefer the enumitem package for lists.”

Instructions are sandboxed: they steer tone, style, and content, but they can’t override the assistant’s tools, its approval rules, or its safety behavior.

  • Keys and host URLs are stored in Oleafly’s local config on your disk. The settings screen says it plainly: keys are stored locally only.
  • Requests go app-to-provider over HTTPS (or app-to-localhost for Ollama).
  • Attachment bytes are never saved into chat history; only file names and types are.

The settings page lists the assistant’s tools with plain-English descriptions, from read_file to compile. The full story, including the approval flow that gates every file change, is on Chat & tools.