Managing Providers
Rephlo connects to multiple AI providers at once. You might use a hosted Rephlo plan for everyday work, your own OpenAI account for general tasks, an Anthropic account for long documents, and a local Ollama model for private data — all configured side by side.
Two Ways to Access AI
1. Rephlo Gateway (Recommended for New Users)
The Rephlo Gateway (the Dedicated API) is the easiest way to get started. Instead of managing your own API keys, you access AI models through your Rephlo account subscription.
Benefits:
- No API key management — just sign in with your Rephlo account.
- Unified billing — one subscription covers multiple AI models, tracked as credits.
- Pre-configured models — popular models available without setup.
- Automatic updates — new models added for you.
To use the Gateway:
- Open the Account area and Sign in with your Rephlo account.
- Once signed in, the Gateway provider is set up for you automatically and its models become available.
The Gateway provider is managed through your account — you don't add or remove it from the provider list manually. See Account and Billing & Subscription.
2. BYOK (Bring Your Own Key)
For users who prefer direct control or already have vendor subscriptions, Rephlo connects directly to AI providers using your own API keys.
Benefits:
- Direct vendor relationship — you pay the provider directly.
- Full control — access every model and feature the vendor offers.
- Custom endpoints — connect to private, enterprise, or self-hosted deployments.
Supported Providers (BYOK)
When you add a provider, you pick one of these types:
| Provider | Vision | Notes |
|---|---|---|
| OpenAI | Yes | GPT-class models; optional custom endpoint |
| Anthropic | Yes | Claude models; supports prompt caching |
| Google AI | Yes | Gemini models |
| Grok (xAI) | Yes | Grok models; optional Live Search (web) |
| Groq | Varies | Fast inference for open models |
| OpenRouter | Varies | Gateway to many models from multiple vendors |
| Ollama | Varies | Local/offline models; optional key for web search |
| OpenAI-Compatible | Varies | Any OpenAI-spec service — Azure OpenAI, vLLM, LocalAI, LiteLLM, custom proxies |
Azure OpenAI is connected through the OpenAI-Compatible type (point it at your Azure endpoint), not as a separate entry in the add-provider list. Ollama is covered in depth on the dedicated On-Device Models page.
Adding a Provider
- Go to Settings > Providers.
- Click Add Provider.
- Select Type — choose the vendor from the dropdown.
- Enter details (fields change based on the type you picked):
- Name — a friendly alias (e.g. "My GPT", "Work Claude").
- API Key — your secret key from the vendor's dashboard.
- Base URL — for OpenAI-Compatible and Ollama, the endpoint URL.
- Model Name — the model to use (required for OpenAI-Compatible and several others).
- Test Connection — validate the configuration before saving (see below).
- Save.
Provider-Specific Fields
Each provider type asks for what it needs:
| Provider | Required | Optional |
|---|---|---|
| OpenAI | API Key | Custom Base URL |
| Anthropic | API Key | Base URL; prompt-caching settings |
| Google AI | API Key | Base URL |
| Grok (xAI) | API Key | Live Search toggle |
| Groq | API Key | — |
| OpenRouter | API Key | Site URL, Site Name (for OpenRouter's activity dashboard) |
| Ollama | Base URL (default http://localhost:11434) | API key (only for web search) |
| OpenAI-Compatible | API Key, Base URL, Model Name | — |
Anthropic Prompt Caching
When you add an Anthropic provider, you'll see prompt-caching options:
- Enable Prompt Caching — on by default. Caching can cut the cost of repeated context by a large margin.
- Cache TTL —
ephemeral(about 5 minutes) orextended(about 1 hour). Longer TTL has a higher write cost but keeps context cached longer. - Cache Logging — optional detailed logging of cache performance.
Caching only kicks in when a request actually carries reusable context (for example, when a Space is attached to a chat). See Prompt Caching for the full picture.
Testing the Connection
The Test Connection button validates your settings before you save:
A green result means the credentials and endpoint work; a red result shows what went wrong so you can correct it.
Model Parameters
Once a model is set, you can tune how it generates text:
- Temperature — lower for precise/deterministic output, higher for creative/varied output.
- Max tokens — caps the length of the response.
- Top-p / sampling — fine-grained control over randomness (where the model supports it).
Rephlo applies parameter ranges that fit each model, so you can't set values a model won't accept.
Multiple Models per Provider
A single provider can hold several models. In the Add/Edit Provider drawer you can:
- Add Model — register another model under the same provider, each with its own parameters.
- Set Active Model — mark which model that provider uses by default.
- Edit / Remove individual models.
This lets one Anthropic or OpenRouter connection cover, say, a fast model for everyday work and a stronger model for hard problems — without configuring the provider twice.
Provider Capabilities
When adding or editing a provider you can flag which capabilities it supports. These drive what's available in Chat and commands:
- Vision — analyze images (screenshots, photos, diagrams).
- Function Calling — call external tools (required for MCP Connectors).
- Long Context — handle large inputs.
- Code Generation — optimized for coding tasks.
- Streaming — responses appear progressively in real time.
For example, you need a Vision-capable provider for image-analysis commands, and a Function Calling-capable model to use connectors in chat.
Setting the Active Provider
You can designate an Active Provider used by default for all commands.
- The active provider runs every command unless a command overrides it.
- To change it, click Activate next to your preferred provider in the list.
- The active provider is highlighted in the interface.
- Only one provider is active at a time.
Getting API Keys (BYOK)
With BYOK you pay the vendor directly for usage. Get keys from each vendor's console:
- OpenAI —
platform.openai.com - Anthropic —
console.anthropic.com - Google AI —
aistudio.google.com - Grok (xAI) —
console.x.ai - Groq —
console.groq.com - OpenRouter —
openrouter.ai - Ollama (web search key) —
ollama.com/settings/keys
Configuring Local Models (Ollama)
To run models entirely offline:
- Install Ollama.
- Pull a model in your terminal, e.g.
ollama pull llama3. - In Rephlo, add an Ollama provider.
- Base URL —
http://localhost:11434(default). - Model — type the name exactly as it appears in Ollama (e.g.
llama3:latest). - (Optional) add an Ollama web-search key to enable web search.
For on-device GGUF models and hardware checks, see On-Device Models.
Provider Interface Features
The Providers page helps you manage many connections:
- Search — find providers by name.
- View toggle — switch between card and list views.
- Status indicators — see at a glance which providers are configured, active, or in error.
Troubleshooting Providers
- 401 Unauthorized — the API key is likely wrong. Re-copy it from the vendor and re-test.
- 429 Too Many Requests — you've hit the vendor's rate limit or run out of credits.
- Connection Refused (Ollama) — make sure the Ollama background service is running.
- Model Not Found — check the model ID is spelled correctly and available on your account.
- Test fails after import on a new computer — encrypted keys can't move between machines; re-enter the key. See Import & Export.
Related: On-Device Models · Prompt Caching · Provider settings
Next Feature: Learn the various ways to Execute Commands.