Settings & providers

Open Settings with the gear at the bottom of the sidebar, beside the project name (or Axon → Settings… on a Mac). Its sections are listed on the left: General, Connectors, Branding, Language, Search, LLM, and Licence where a licence applies.

#What runs where

Axon ships with local models — chat, embeddings, and OCR (for scanned documents) — that run inside the app, on your machine, covered by the ~3.8 GB download from first launch. Axon picks the size of chat model that fits your machine. GPU acceleration is automatic where available (Metal on Apple Silicon; Vulkan or CUDA on other machines when the build includes it), and falls back to CPU otherwise. Nothing about the local models ever leaves your machine.

You're never required to add anything beyond that. Everything below is optional.

#Providers

Under Settings → LLM, Axon supports:

• Local — the bundled model, sized to your machine. On this machine shows the local chat model, whether the models are downloaded, and the GPUs it uses.
• Ollama — your own Ollama server, if you run one. The Ollama section shows whether it's installed and running, with Start Ollama or Get Ollama; once it's running, its models appear in the model picker automatically. It's not required. See How to: Connect Axon to your own Ollama server.

Both run on hardware you control and need no account or key: Axon has no cloud model providers, so your conversations and documents are never sent to one. Choose which model answers with the model picker in the composer — it's not a project setting, so different chats can use different models.

#Models

Settings → LLM → Models shows the models behind search and indexing:

• Embedding Model builds your search index. Changing it is a bigger decision than it looks: everything already indexed was embedded with the old model, so the affected projects are re-indexed — once, in the background — after a change like this.
• OCR Model reads scanned PDFs and images.
• Summary model writes each document's summary at index time — Same as chat model unless you choose another.
• Context window (local models) — how much the local model can read at once. Automatic sizes it to your machine.
• Evidence limit — how close a passage must be to your question to be used as evidence. Lower refuses more and answers less.

#General

• Appearance — Theme (light or dark), and Show skills, which adds the Skills section to the sidebar (see Skills).
• Projects — where new projects are created. See How to: Change where new projects are created.
• Setup — Setup assistant → Open reopens the first-run setup, to download or re-download the local models.

#Language

Interface language sets the language of Axon's own buttons and labels, and Response language the language the assistant replies in. Dictation chooses the Speech model for voice input and the Spoken language.

#Search (connectors)

The composer's Web search is separate from chat — it's how you find pages to review and add as sources, not something the model reaches for mid-answer. Choose a search engine (Brave Search API, or DuckDuckGo without a key, though DuckDuckGo often refuses automated requests), and switch on free academic and reference sources, under Settings → Search. See How to: Set up web search for connectors.

#Connectors

Under Settings → Connectors you connect the accounts a chat can search: OneDrive today, with Google Drive and SharePoint marked Coming soon. Connecting an account is global — once it's connected, every project can search it. In a chat, choose it under Search in the composer's + menu; results come back for you to review, and nothing becomes a source until you Add to sources. Web is always there and needs no account. Searching a cloud drive is a Pro feature; web search is not.

#Branding

Under Settings → Branding you set the letterhead your documents carry: your logo, firm name, an accent colour and a footer line. You can fill it in on any plan; documents carry it once Pro is active. See Artifacts & documents for what each kind of file gets.

#Licence

Settings → Licence is where you get and activate Pro. Without it, Get Pro opens the plans on exaflux.io. Once you've bought it, Connect this computer opens exaflux.io in your browser — sign in and approve the code shown in Axon; there's no key to copy. Each computer is connected separately. Under Pro includes it lists what a licence adds:

• Sharing on your network — sharing a project with another computer and keeping it in sync. The computer that takes the copy needs no licence.
• Cloud-drive search — searching OneDrive (and Google Drive and SharePoint, once available) from a chat.
• Branded exports — your letterhead on Word, PDF, PowerPoint and Excel files.

Everything else — the local models and Ollama, web search, citations, audit records and every export — works on any plan. A Pro badge anywhere in the app leads to this tab.

#Where things live

Settings and your projects list all live in files on your machine — nothing is tied to an account beyond an optional Pro licence. Help → Show Logs and Copy details in the About window are the fastest way to get diagnostic information; see Troubleshooting.