Skip to content

Frequently Asked Questions

API & Providers

How do I get an API key?

Can I use multiple API keys for the same provider?

Yes. Each provider supports multiple named keys. Tap the radio button to select the active key. Useful for rotating between work/personal keys or having a backup ready. See API Providers.

How do I add a custom provider?

Go to Settings → Provider → + Add Custom Provider. Enter a name and base URL. Any OpenAI-compatible endpoint works. See Custom Providers.

Local Models

What GGUF models work?

Agora supports GGUF format for both chat and embedding. Chat models should fit in device memory (1–8B parameters depending on RAM). Embedding models are much smaller (100–500 MB). See Local Models.

How do I run models offline?

Import a GGUF chat model via Settings → Provider → Local → Import GGUF Model. For fully offline semantic search, also import a GGUF embedding model. No network connection needed.

Why is my local model so slow?

Local inference runs on your device's CPU. It's inherently slower than cloud APIs. Tips: use smaller models (1–3B parameters), lower quantization (Q4_K_M), shorter context windows, and close background apps.

Why is my embedding model test failing?

Common causes:

  • Wrong model name — check exact spelling, including Ollama tags (e.g., qwen3-embedding:8b not qwen3-embedding)
  • Wrong base URL — ensure the endpoint supports /v1/embeddings
  • Missing API key — some providers require authentication even for embeddings
  • Network — check connectivity to the endpoint

Keyword search matches exact text. RAG (semantic search) matches by meaning — "database setup" can find "Room configuration" even without shared words. RAG requires an embedding model and cached messages. See Conversation Search.

How do I use Ollama for embeddings?

  1. Install Ollama on a machine
  2. Pull an embedding model: ollama pull qwen3-embedding:8b
  3. In Agora, add a remote embedding model with the Ollama preset
  4. Use http://<host>:11434/v1 as base URL
  5. Enter the exact model name including the tag (e.g., qwen3-embedding:8b)
  6. Leave API key blank

Memory

What's the difference between Active Memory and Saved Memories?

Active Memory is a single persistent context included with every API call — the model always sees it. Saved Memories are a collection of named files the model searches and retrieves on demand. Use Active Memory for persistent facts; use Saved Memories for reference material. See Memory & Cache.

Can the model modify my memories?

Yes, if you enable Access Saved Memories and/or Access Active Memory in Settings → Memory. The model can create, read, edit, and delete memories via tool calls. All permissions default to off.

Shell & Tools

How do I set up remote shell access?

Deploy the Conch server on your target machine, then add the device in Settings → Shell with its URL and API key. See Remote Shell.

Is the shell connection encrypted?

Yes. Conch uses ECDH key exchange + AES-256-GCM encryption + HMAC-SHA256 signing. All traffic between Agora and the Conch server is end-to-end encrypted.

Data

How do I back up my data?

Go to Settings → Data Control → Export Data. Select categories and export to a .agora file. Store it somewhere safe. See Data Portability.

Can I import from ChatGPT or Claude?

Yes. Export your data from ChatGPT or Claude (they provide .zip files), then import in Settings → Data Control → Third Party. Both Merge and Replace strategies are supported. See Data Portability.

Are my API keys included in exports?

They can be, but it's optional. The export screen lets you toggle API key inclusion. A warning is shown when you enable it. Keys are stored in plain text within the .agora file, so only include them for full device migrations to trusted destinations.

General

Where is my data stored?

Everything is stored locally on your Android device in a Room database. Agora has no servers, no cloud sync, no telemetry. Messages are sent directly from your device to the AI provider you configure.

Does Agora support multiple languages?

Yes. The app UI supports English and 中文 (Chinese). Settings → Language. A restart is required after switching.

How do I report a bug or request a feature?

Open an issue on GitHub. For contributions, see the Contributing section of the README.