DocumentationAI
Connect an AI Provider
Configure and verify AI language models for automated summaries, action items, custom templates, and meeting chat.
Transcription and AI are architecturally independent in Kapinote:
- Speech Transcription: Transcribes streaming audio into time-stamped text segments;
- AI Models: Synthesize transcripts into structured summaries, extract action items, and answer interactive questions about your meetings.
Adding a provider
- Open Settings → AI.
- Select Cloud, Local, or CLI, then choose your specific provider.
- Enter the required fields (API key, model identifier, or custom Base URL).
- Click Verify to validate network connectivity and authentication.
- Save the configuration and set it as your active AI provider.
Configuration fields reference
| Field | Description & Syntax |
|---|---|
| API Key | Developer API credential generated in the provider's console (not your personal account password) |
| Model | Case-sensitive official model identifier (e.g., claude-3-7-sonnet-20250219, gpt-4o, deepseek-chat) |
| Base URL | API endpoint. Keep Kapinote's default unless your organization routes through a custom proxy or private deployment |
| Region | AWS Bedrock uses the Base URL field to specify the AWS Region identifier (e.g., us-east-1) |
For CLI providers, you can leave the Model field blank to automatically inherit that CLI's configured default model.
Choosing the right integration type
- Cloud APIs: Quickest setup, high inference speed, and access to flagship frontier models (Claude 3.7, GPT-4o, DeepSeek-V3, Gemini 2.5). Requires sending transcript context to the provider API.
- Local with Ollama: Requires no API key and can process meeting text on the Mac after the model is downloaded. Speed and quality depend on model size, unified memory, and chip performance (e.g.,
qwen3.5:4b,qwen3.5:9b,qwen3.8:27b). - CLI Tools: Reuses existing terminal logins from Gemini CLI, Claude Code, GitHub Copilot CLI, or OpenAI Codex CLI with zero API key entry. Billed under your existing tool subscription.
See Cloud AI providers, Local AI with Ollama, or Local and CLI providers for detailed walkthroughs.
Troubleshooting verification failures
- Model Identifier: Verify case-sensitive spelling against official provider documentation.
- Quota & Balance: Ensure your API key has access to the specified model and sufficient credit balance.
- CLI Permissions: Test the command directly in Terminal to confirm valid login before verifying in Kapinote.
- For additional solutions, see Troubleshooting.