Custom model setup

Flowix ships an "Agent CLI" abstraction layer, but if you don't want to install a CLI, you can just configure an OpenAI-compatible HTTP endpoint.

This path covers: local Ollama / LM Studio / Jan.ai, as well as proxy services for regional models (DeepSeek, Zhipu, Moonshot, etc.).

1. General setup

Open Flowix Settings > AI > Custom, click + New endpoint, and fill in:

Field	Required	Notes
`Name`	✓	Shown in the Agent dropdown, e.g. `Local Ollama`
`Base URL`	✓	Something like `http://localhost:11434/v1`
`Model`	✓	Model ID, decided by the server
`API key`	-	If the server doesn't verify, fill `EMPTY`
`Stream`	✓	Usually on; long output flows more smoothly
`Max tokens`	-	Server cap; leave blank for default
`System prompt template`	-	Prepended to every call, see below

Save and you're done. Flowix POSTs to {base_url}/chat/completions, fully following the OpenAI Chat Completion API spec.

2. Common configurations

Ollama

Name:     Local Ollama
Base URL: http://localhost:11434/v1
Model:    qwen2.5:14b
API key:  EMPTY

Note: requires ollama serve running in the background.

LM Studio

Name:     LM Studio
Base URL: http://localhost:1234/v1
Model:    qwen2.5-coder-32b-instruct
API key:  lm-studio

DeepSeek official

Name:     DeepSeek
Base URL: https://api.deepseek.com/v1
Model:    deepseek-chat
API key:  sk-...

Zhipu GLM

Name:     Zhipu GLM-4
Base URL: https://open.bigmodel.cn/api/paas/v4
Model:    glm-4-plus
API key:  your API key

Note: Zhipu's Base URL doesn't end with /v1; remember that Flowix automatically appends /chat/completions.

3. System prompt template

For every document call, you can set a global prefix on your endpoint -- ideal for behavior shaping:

You are running inside Flowix (a local-first Markdown notes app).
Always answer in Simplified Chinese; wrap code in ```code blocks```.
Do not explain Markdown syntax itself.

This works alongside document-level prompts -- the document-level prompt is written near the cursor, the system-level prompt lives in Settings; the latter takes effect first.

4. Differences from the Hermes path

	Custom endpoint	Hermes
Protocol	HTTP	Calls hermes-cli
Tool calls	Depends on model	Built-in
Multi-step Agent	No	Yes
Best for	Simple "Q&A" continuation	Complex multi-file tasks

Use a custom endpoint for simple writing / translation / explanation. For code edits, running commands, and multi-file editing, stick with Claude Code or Codex.

5. Debugging

Under Settings > AI > Custom > [endpoint] > Debug, enabling it will dump the full request / response to ~/Library/Logs/Flowix/custom.log (macOS) on the next call:

{
  "request": { "model": "qwen2.5:14b", "messages": [...] },
  "response": { "choices": [{ "message": { "content": "..." } }] },
  "ms": 1820
}

Useful for troubleshooting timeouts and quota errors.

6. Security reminders

Don't commit production keys to a git repo -- even private ones leak via git backups
When a corporate proxy goes through the system proxy, use http://127.0.0.1:7890 in Base URL; don't hardcode the real address
Delete endpoints you no longer use to avoid accidental reuse later