Quick start

Install, connect, work as usual.

For people who do not use a terminal. Buttons are named exactly as the console shows them.

Install

Windows

  1. Download FULI-Setup-<version>.exe, double-click it and click “Install”.
  2. When “Windows protected your PC” appears, click “More info”, then “Run anyway”.
  3. No administrator permission, and no Node.js, Java or database to install.

macOS (Apple silicon)

  1. Download fuli-macos-arm64-<version>.dmg and drag FULI.app to Applications.
  2. Open FULI from Applications, not from the disk image.
  3. If it is blocked the first time, click “Open Anyway” in System Settings › Privacy & Security.

Needs macOS 13.5 or later. Intel Macs use the terminal install.

Open FULI: it starts by itself and opens the console in its window. If not, click “Start FULI”.

Connect your AI clients

  1. The console first shows “Get started with FULI”: tick the clients it found and click “Connect all”.
  2. Then answer “Where should FULI record?” and click “Start”. Nothing is recorded until you choose.
  3. Later, use Settings › Clients: “Connect” or “Connect all”; after an update, “Update connection”.
  4. A client not listed: click “Copy MCP configuration” and paste it into its MCP settings.
  5. Restart the client afterwards. Sessions already open need a restart to load the new FULI.
  6. Using Codex: type /hooks in Codex and trust FULI’s hooks.

“Connected” only means the configuration is there, not that the client is connected.

Work as usual

  1. In your project folder, give your AI tasks as you always do. No need to call FULI.
  2. On every prompt, FULI hands the AI the preferences and knowledge that apply.
  3. Conclusions worth keeping are saved. The reply ends with what this turn saved, with links to edit.
  4. Saved knowledge is in Memory › Knowledge; what needs your confirmation, in Memory › Organize.
  5. Anything the AI waits for you to decide is under “Decisions” in the top bar.
  6. Not sure what to say? Settings › How-to has phrases you can copy.

Optional: connect a model

FULI works without one: the AI answers in the current session and FULI matches words.

The model button

  1. Beside Settings in the top bar. Not connected: a grey dashed ring. Connected: a glowing gold bolt.
  2. While old knowledge is being indexed, a ring around it fills up.
  3. It opens “Models and usage”. Each card under “Uses” says what model is enough; click “Connect”.

Find by meaning: semantic search

  1. On “Semantic search”, click “Connect”, choose “Local Ollama” and pick or enter qwen3-embedding:0.6b.
  2. Or pick “Online service”: choose a suggested endpoint (such as Qwen) and a model from its list.
  3. Turn on “Use the model set below” and save. “Details” has “Test connection”.

Ollama has to be installed on this computer, with the model downloaded.

Older knowledge: retrieval terms

  1. Connect a model to Astro, then turn on “Allow adding retrieval terms to older knowledge” there.
  2. Pick a project and read the free estimate: how many items lack terms, and about how many tokens.
  3. Click “Try 20 first” to compare before and after, then “Fill the other N” or “Skip”.

You can click “Stop” at any time while it fills.

When something is wrong

  1. FULI checks itself. If the top bar says “N things need you”, click “View”, then “Fix automatically”.
  2. Windows: if FULI did not start because its port is reserved, choose “Use another port” in the tray.
  3. A dot beside the version in the top bar means an update: click the version, then “Update to v…”.
  4. Updates keep your knowledge. Afterwards, reopen any AI sessions that were open.
  5. For the developers: Settings › System, “Create diagnostics bundle”, then “Export”. No knowledge inside.