Help
The usual problems, and what to do. If yours is not here, open an issue.
Installing and starting#
| Problem | What to do |
|---|---|
totem says it could not start | It says why. The port in use: use another one (totem 4524). No permission to write the data: choose a folder of yours (totem --data <folder>). A broken installation: install it again. The whole output of the server is in totem.log in the data folder |
totem says it is already running | It opens the one that runs. totem status says where it listens and totem stop stops it |
I started it with totem serve and the terminal is busy | That command keeps Totem in the terminal until Ctrl+C. Use totem to get the terminal back |
| Where is my data? | totem config prints the folder |
| The container does not start | It needs an access token: mount a file at /run/secrets/totem_token, or set token in config |
| Docker cannot see my folder (Colima, Docker Desktop) | Share the folder with the virtual machine: your home folder, not /tmp |
Models#
| Problem | What to do |
|---|---|
| There is no model to choose | Settings, then add a provider, or Look for servers on this computer with Ollama or LM Studio running |
| A small model seems to ignore the instructions | Check ollamactx in config: Ollama must get a real context (Totem asks for 16384 by default) |
| A model repeats itself until it stops | Totem asks Ollama to penalize repeating and to stop one answer after a limit (ollamarepeat, ollamapredict, and a lower ollamapredictsmall for models up to 4B) |
| The model answers with code in spaces and not tabs | Totem checks the code of the answer and asks the model to correct it once. A very small model does not always manage |
| A model does not call tools | It does not support them: Totem sends it the request without tools. Choose another model for editing |
Total.js#
| Problem | What to do |
|---|---|
| The agent asks to start Forge | It is the connection to the documentation. Allow it once for every project, or set forge : false |
| The agent mentions jComponent in server code | Totem's recipes and its reminder keep the interface out of server questions, but a small model may slip. Tell it; or use a bigger one |
| LiteLLM in front of Ollama: the model answers with JSON, or forgets what I asked | In the config of LiteLLM write the model as ollama_chat/<model> (not ollama/). Totem sends the window (num_ctx) by itself |
| I do not want the recipes or the code check | recipes : false and selfcheck : false in config |
| I want Forge only through its MCP server, or not at all | forgedocs : false (MCP only, asks to start) or forge : false; forgemcp : true adds the MCP tools to the documents |
The interface#
| Problem | What to do |
|---|---|
| "The terminal could not start" | The Shell needs Python 3 (macOS and Linux). It is in the Docker image |
| A worktree could not be created | The project needs git with at least one commit |
| Scheduled prompts do nothing | They run unattended: in Plan mode they only read, and anything that would ask is refused. Choose Build and auto-accept to let them edit |
| A scheduled prompt does not start | A schedule never starts while its previous run is still going. Check the model is available and the folder exists |
| The session says "The worktree of this session was discarded" | Start a new session: a discarded worktree cannot run |
Security#
| Problem | What to do |
|---|---|
| Can I open Totem to my network? | Yes, with a token: $ip in config and token. Without a token it refuses to start. See security |
| Does the model see my keys? | No. Keys and the token are scrubbed from everything the model sees, and .env files ask first |