Skip to content

API Providers

Agora connects directly to AI providers — no middleman, no subscription, no telemetry. You bring your own API keys, and everything runs from your device.

Built-in Providers

Provider Base URL Models Notes
Google https://generativelanguage.googleapis.com/v1beta Gemini series Free tier available via Google AI Studio
OpenAI https://api.openai.com/v1 GPT-4, GPT-4o, o-series Reasoning models supported
Anthropic https://api.anthropic.com/v1 Claude series Extended thinking supported
DeepSeek https://api.deepseek.com/v1 DeepSeek-V3, DeepSeek-R1 Reasoning models supported
Qwen https://dashscope-intl.aliyuncs.com/compatible-mode/v1 Qwen series Via Alibaba DashScope
Ollama http://localhost:11434/v1 Any pulled model Self-hosted, no API key needed
OpenRouter https://openrouter.ai/api/v1 Multi-provider Access many models through one API
Local N/A GGUF models On-device via llama.cpp, fully offline

Switching Providers

Tap the provider selector in Settings to switch between providers. Each provider maintains its own:

  • API keys
  • Base URL (editable for proxies/self-hosted)
  • Model list

API Keys

Multiple Keys per Provider

Each provider supports multiple named API keys. This enables:

  • Rotation — switch between keys for different usage tiers
  • Organization — separate work and personal usage
  • Fallback — keep a backup key ready

Managing Keys

  1. Go to Settings → Provider
  2. Select a provider
  3. Under API Keys, tap Add New Key
  4. Enter a name (e.g., "Work", "Personal", "Team Shared") and the key value
  5. Tap Add

Tap the radio button to set the active key. Long-press a key to Edit or Delete.

Key Safety

Warning

API keys are stored locally in an encrypted Room database. They are never sent to Agora servers (there are none). However, they are exported in plain text if you include them in a .agora export file.

Custom Providers

Add any OpenAI-compatible API endpoint:

  1. Go to Settings → Provider
  2. Tap + Add Custom Provider at the bottom of the provider list
  3. Enter:
    • Provider Name — any display name
    • Base URL — the API endpoint
  4. Tap Add

Agora fetches the model list from {base_url}/v1/models. Once added, custom providers work exactly like built-in ones: add API keys, sync models, and chat.

Use Cases

  • Self-hosted — connect to vLLM, LocalAI, text-generation-webui, or other OpenAI-compatible servers
  • Proxies — route through a corporate proxy or API gateway
  • Alternative endpoints — use Azure OpenAI, Cloudflare AI Gateway, or other compatible services

Rename or Delete

Long-press a custom provider to Rename or Delete. Deleting removes the provider and all its keys.

Warning

Built-in providers cannot be renamed or deleted.

Base URL Override

Every provider (including built-in) has an editable Base URL. This is useful for:

  • Proxies: Route through https://my-proxy.example.com/v1
  • Self-hosted: Point to your own instance
  • Region routing: Use region-specific endpoints

Syncing Models

After adding API keys, sync the model list:

  1. Go to Settings → Models
  2. Tap Sync from All Providers
  3. Agora fetches available models from every configured provider

A snackbar shows sync progress and results. You can then enable/disable individual models and set a default.

Provider-Specific Notes

Google Gemini

  • API keys from Google AI Studio
  • Free tier available with rate limits
  • Supports code execution and search grounding (built-in tools)

OpenAI

  • API keys from Platform
  • Reasoning models (o1, o3) require specific API access
  • Streaming, tools, and vision all supported

Anthropic

  • API keys from Console
  • Extended thinking with configurable token budgets
  • Tool use with parallel calls supported

Ollama

  • No API key required (local network)
  • Base URL typically http://<host>:11434/v1
  • Model list fetched from Ollama's API
  • See FAQ for Ollama-specific troubleshooting

OpenRouter

  • Single API key for 200+ models
  • Pay-per-token pricing varies by model
  • Good for trying different models without individual provider accounts

Local (llama.cpp)

  • No network required
  • GGUF model files stored on-device
  • See Local Models for setup