Find the message you see, then try the fix next to it. Messages are shortened; … stands for the part that changes.
Setup and connections
| You see | Try |
|---|
No Gemma model configured | Set ollama.model to a name from ollama list. |
Ollama is not reachable | Start Ollama (ollama serve or the app). |
Tier 1 runs offline: the Ollama host must be http on loopback | ai-coach only talks to Ollama on your own machine. Point ollama.host at 127.0.0.1 or localhost. |
Could not start the vault server | Check docsMcpBin points at kb-template's docs-mcp.js, and that kb-template is built (pnpm run build). The error includes the end of the server's log. |
unexpected response from the vault server | Your kb-template build and ai-coach disagree on a tool's reply. Pull and rebuild kb-template. |
config.json: … with a setting name | The file has a typo or an unknown key. The message names it. |
Sparring and synthesizing
| You see | Try |
|---|
This note needs about N tokens, but Gemma's context window is … | Raise ollama.contextWindow, or spar a shorter note. |
only notes under 3.Resources can be sparred | File the note into Resources first: coach file <note> 3.Resources/<topic>. |
… is raw material: spar a concept note that quotes it instead | Book chapters are source material. Run coach synthesize to make a concept note. |
Nothing marked in … yet | Bold or highlight some passages, or write a chapter summary, then synthesize again. |
A concept note needs at least one core assertion | Keep at least one of the drafted assertions, or type your own. |
The judge
| You see | Try |
|---|
| A quota or server error | Try again later, or set judge.fallbackModel. Your sessions stay queued until judged. |
| An unknown model | Set judge.model and judge.fallbackModel to models your key has. |
Still stuck?
Small local models make mistakes. Every reply is checked and retried once, and ollama.checkModel makes the end-of-round verdict fairer, but a sparring verdict is a first opinion. If it looks wrong, run coach judge for a second one.