Configuration

Provider, model, voice, hotkey, privacy, diagnostics, and persisted defaults for FlowLens.

Runtime storage model

FlowLens has two layers of configuration:

Layer	Purpose	Storage
Non-secret preferences	hotkey, mode, provider base URL, model, voice ID, mic, overlay position, privacy, updates	`app.getPath('userData')/config.json`
Secrets	provider API key and ElevenLabs API key	encrypted `app.getPath('userData')/secrets.json`
Legacy migration	older development config	one-time copy from `~/.flowlens/config.json`

In tests and lower-level config helpers, the legacy path can still be used directly. In the running Electron app, runtime-config.ts routes config through userData.

Reference shape

{
  "hotkey": "CommandOrControl+Shift+Space",
  "defaultMode": "error-explainer",
  "firstRunComplete": false,
  "provider": "openai",
  "providerKey": "openai",
  "providerProtocol": "auto",
  "providerBaseUrl": "https://api.openai.com/v1",
  "model": "gpt-4o-mini",
  "apiKeys": {},
  "tts": {
    "model": "eleven_flash_v2_5",
    "voiceId": "JBFqnCBsd6RMkjVDRZzb"
  },
  "audio": {
    "inputDeviceId": null
  },
  "overlay": {
    "position": { "mode": "bottom-right" }
  },
  "launchAtLogin": false,
  "updates": {
    "channel": "stable",
    "autoDownload": false
  },
  "privacy": {
    "confirmBeforeScreenshot": false,
    "voicePlaybackEnabled": true
  },
  "diagnostics": {
    "loggingEnabled": true
  }
}

Field guide

Field	Purpose	Default
`provider`	Adapter family. Current implementation uses the OpenAI-compatible adapter.	`openai`
`providerKey`	Secret-store key for the active provider credential.	`openai`
`providerProtocol`	Wire protocol for compatible endpoints. `auto` lets FlowLens pick from known provider presets.	`auto`
`providerBaseUrl`	Base URL for the OpenAI-compatible API.	`https://api.openai.com/v1`
`model`	Model passed to `/chat/completions`.	`gpt-4o-mini`
`tts.model`	ElevenLabs text-to-speech model.	`eleven_flash_v2_5`
`tts.voiceId`	ElevenLabs voice ID selected by setup/settings.	`JBFqnCBsd6RMkjVDRZzb`
`audio.inputDeviceId`	Browser microphone device ID. `null` means system default.	`null`
`overlay.position`	Bottom-right anchor or custom dragged position.	`bottom-right`
`launchAtLogin`	Starts FlowLens in the background after OS login.	`false`
`privacy.voicePlaybackEnabled`	Enables or disables spoken TTS playback.	`true`
`diagnostics.loggingEnabled`	User-facing preference for diagnostics logging.	`true`

Provider configuration

FlowLens supports providers through one OpenAI-compatible adapter plus provider-specific presets where needed:

providerKey selects which stored secret to use.
providerBaseUrl selects the API root.
providerProtocol selects or infers the transport.
model selects the model.

For normal OpenAI-compatible providers, the request URL is assembled as:

{providerBaseUrl}/chat/completions

The current setup/settings UI also includes convenience presets:

Provider path	Base URL	Protocol	Notes
OpenAI-compatible	`https://api.openai.com/v1`	`openai-chat`	Default adapter path for vision-capable chat completions
Gemini OpenAI-compatible	`https://generativelanguage.googleapis.com/v1beta/openai`	`openai-chat`	Works through Google's OpenAI-compatible chat completions surface
OpenCode Go	`https://opencode.ai/zen/go/v1`	`auto`	Infers per-model protocol and capability quirks
Custom compatible endpoint	user supplied	user selected or `auto`	Useful for local gateways or hosted OpenAI-compatible APIs

OpenCode Go has extra compatibility handling because its available models do not all behave like a single strict OpenAI clone. FlowLens can omit native response_format where the endpoint rejects it, infer text-only versus vision-capable models, avoid the short app-side request timeout for long-running models, and coerce prompt-only markdown output into the structured response contract when the answer is useful but not valid JSON.

This keeps the overlay, prompts, response parser, and voice pipeline stable while still allowing provider-specific request differences.

ElevenLabs configuration

FlowLens uses ElevenLabs for both ends of the voice loop:

scribe_v2 for speech-to-text
configurable TTS for spoken summaries

The current settings UI exposes these TTS model options:

Model	Use
`eleven_flash_v2_5`	Lowest latency for quick overlay responses
`eleven_turbo_v2_5`	Balanced speed and quality
`eleven_v3`	More expressive speech
`eleven_multilingual_v2`	Stable multilingual playback

The voice picker loads available voices through the main process using the stored ElevenLabs key. The renderer never receives that key.

Settings and cleanup

The settings window and in-overlay settings panel expose:

provider key, base URL, model, and API key
ElevenLabs key, TTS model, and voice picker
microphone selector
hotkey capture and reset
permission checks and request actions
overlay position reset
launch-at-login
voice playback
screenshot confirmation preference
diagnostics export
factory reset

Factory reset clears FlowLens-owned settings, encrypted secrets, diagnostics logs, and overlay position.