Ai Keeper/Guides/First steps
← All workflows
Collection 01 · Start safely

First steps

Go from a downloaded app to a working local chat without learning model formats, ports, or inference engines first.

5 workflowsBeginner-first

Find one workflow

Filter this collection by name, outcome, provider, engine, channel, mode, or command.

5 shown
Workflow 01

Install, open, and complete first-run setup

Move the app into Applications, complete macOS safety prompts, and use the two-step Server or Client setup.

Beginner3–5 minResult · The correct operating mode opens✓ Source-audited workflow
Current Ai Keeper Home hub as it appears after first launch.
This is the screen you should reach before configuring anything. If you land somewhere else, the app restored a previous session.
Before you start
  • An Apple Silicon Mac running macOS 14 or later.
  • The Ai Keeper disk image downloaded from the official site.

  1. Open the downloaded .dmg, then drag Ai Keeper into Applications.

    Running from Applications lets updates, permissions, and managed engine paths behave predictably.

  2. Open Applications in Finder and double-click Ai Keeper. If macOS asks whether you trust the app, confirm only if the publisher and download source are the ones you expect.

  3. Allow notifications only if you want completion and automation alerts. Microphone, speech, camera, and screen-recording access can wait until a workflow needs them.

  4. On Welcome, choose Server when this Mac will run models, or Client when it will control another Ai Keeper host.

  5. In Server mode, use Start Chat to accept the recommended first path: one engine, one model, then Chat. Choose the optional Work Mode wizard only when you want its five-step configuration.

  6. In Client mode, enter the remote management address and key on Connect. Finish setup only after reachability succeeds.

You are done when
Server mode reaches the 13-hub Home window; Client mode reaches the seven-hub client window with Administration instead of local model-management hubs.
If something looks wrong
macOS says the app cannot be openedDownload it again from the official source. Do not bypass a signature warning for an unknown copy.
A permission was deniedThat is fine for first launch. The relevant Voice, Camera, or Screen Capture panel can request it again later.
Workflow 02

Your first 30 minutes

Run the health check, install the required engine pieces, download a recommended model, start it, and send one message.

Beginner20–30 minResult · One local model answers in Chat✓ Source-audited workflow
Current Ai Keeper Home hub showing one Ready local route and the primary navigation.
Home is the safest checkpoint: Ready, Online, Processes, and Attention summarize whether you can continue.
Before you start
  • Ai Keeper is open.
  • A stable internet connection for the first engine and model download.
  • At least 10 GB of free disk space; larger models need much more.

  1. Open Activity > Health and run Diagnostics or Doctor. Read any red item before continuing.

    A missing command-line tool is much easier to fix before a multi-gigabyte model download.

  2. Open Engines and click Fix Required. Let the operation finish; use Output Log only if a row reports an error.

  3. Open Settings, search for Storage, and confirm the models folder is on a disk with enough free space.

  4. Open Models > Downloads, find a small Apple-Silicon-compatible chat model, and download it. Review its findings under Trust before first launch.

  5. When the download finishes, open Runtime > Instances. Select the new instance and click Start. Leave the engine on Automatic.

  6. Wait until the instance says Ready. Open Chat, select that model, type Reply with exactly: local model ready, and send.

You are done when
Chat streams a reply from the selected local instance, and Runtime still shows that instance as Ready.
If something looks wrong
The model does not fitReturn to Models > Downloads and choose a smaller quantization.
The instance never becomes ReadyOpen its detail panel and run Diagnose. Then check Engines and Activity > Logs instead of repeatedly clicking Start.
Chat has no model to selectThe instance must be Ready and have a Chat/LLM role.
Workflow 03

Choose Server mode or Client mode

Decide whether this Mac runs the models or controls another Ai Keeper Mac.

Beginner3 minResult · The app uses the correct operating mode✓ Source-audited workflow
Settings Connection showing Server and Client modes and the masked management API key.
Use Server on the Mac that runs models; choose Client only when this Mac should connect to an existing Ai Keeper server.
Before you start
  • Use Server mode for a standalone Mac.
  • For Client mode, another Ai Keeper Mac must already be running as a reachable server.

  1. Open Settings > Connection.

  2. Choose Server if this Mac will download engines, hold models, and run inference. This is the normal first-Mac choice.

  3. Choose Client only if another Mac is the workhorse. Enter that server's management address and management API key.

  4. Save, then return to Home and wait for the connection summary to refresh.

You are done when
Server mode shows local runtime controls. Client mode reports the remote host as reachable and intentionally hides local-only Downloads, Models, Engines, and Dependencies surfaces.
If something looks wrong
Client mode looks emptyConfirm the remote server address includes the correct scheme and port, and that its management API key matches.
You only own one MacUse Server mode. Client mode is not a lightweight local mode.
Workflow 04

Apply a Work Mode

Use an outcome-based preset to configure sensible model, memory, tool, trust, and automation defaults.

Beginner5–10 minResult · A saved mode is active✓ Source-audited workflow
Settings Work Modes showing the current setup and outcome-driven presets.
Start from a recommended card, then review its defaults and requirements before applying it.
Before you start
  • At least one Ready chat model is helpful.
  • Some modes need optional packages; the mode card will show them.

  1. Open Settings, search for Work Modes, and pick the outcome closest to your job, such as research, coding, writing, or private local work.

  2. Read the mode summary, especially its autonomy, web, memory, and tool choices.

  3. Open the mode's requirements. Install or resolve only items marked missing; do not add every optional package.

  4. Click Apply. Review any confirmation describing settings the mode will change.

  5. Open the destination suggested by the mode and run its starter task.

You are done when
The mode is marked active or applied, its health shows no blocking requirement, and the starter surface opens with the expected defaults.
If something looks wrong
A requirement is blockedUse the requirement's Review or Install action; it takes you to the exact Engines, MCP, ClawHub, or Secrets surface.
You edited the modeUse Reapply when you want the saved definition to overwrite the current drift.
Workflow 05

Find your way around the current hubs

Learn the one-sentence job of every primary destination so you always know where to start.

Beginner5 minResult · You can locate any feature✓ Source-audited workflow
Current Ai Keeper Home hub with the 13 Server-mode destinations visible in grouped navigation.
Use the sidebar as the map and the Home shortcuts as the fastest routes into common workflows.
Before you start
  • Ai Keeper is open.

  1. Use Home for status; Models to download and review model trust; Runtime to run models; and Engines to install their software.

  2. Use Chat for conversations plus Assistant, Voice, Browser, and Canvas tools; use Media for local images, video, music, and enhancement.

  3. Use Knowledge for context and sources, Agents for agents/workspaces/runs, and Automation for scheduled or event-driven work.

  4. Use Extensions for channels, plugins, MCP, skills, and ClawHub; Lab for API, QA, transcription, and Harness experiments; Activity for traffic and diagnostics; and Settings for configuration and operator tools.

  5. Press ⌘K in this manual whenever a label is unfamiliar.

You are done when
You can answer three questions without guessing: where to start a downloaded model (Runtime), where to index a PDF (Knowledge), and where to inspect a failed request (Activity).

Keep going

← All workflowsBrowse the complete libraryEvery Work Mode →Next collection