Add Your First Provider

A provider is the company supplying the LLM service — OpenAI, Anthropic, Google, or any service that exposes an "OpenAI-compatible" endpoint. ETOS LLM Studio ships with no built-in model; every conversation goes through a provider you add.

This page walks through adding your first provider in detail and confirming text chat works. Advanced configuration (header overrides, parameter expressions, raw JSON body) lives at the bottom — skip it on your first pass.

Read This First

Three Things to Have Ready

Need	Where to get it
A working API key	From the vendor console. OpenAI: platform.openai.com/api-keys. Anthropic: console.anthropic.com. Google: aistudio.google.com/apikey.
The Base URL	Official vendors: see the table below. Third-party relays: look for "OpenAI compatible URL" or base_url in their docs.
At least one valid model ID	e.g. gpt-4o, claude-3-7-sonnet-latest, gemini-2.5-pro. Model IDs are exact strings — a single typo means "model not found".

One Provider Is Enough at First

Configure just one stable provider on your first pass. Add a second only after chat is working.

Official Base URL Cheat Sheet

All addresses below are base URLs — fill in only to /v1 (or the equivalent version segment). The app appends the actual endpoint path (/chat/completions, /messages, etc.) automatically.

Vendor	API Format	Base URL
OpenAI (Chat Completions endpoint)	OpenAI Compatible	https://api.openai.com/v1
OpenAI (Responses endpoint, for GPT-4.1 / GPT-5)	OpenAI Responses	https://api.openai.com/v1
Anthropic Claude	Anthropic	https://api.anthropic.com/v1
Google Gemini	Gemini	https://generativelanguage.googleapis.com/v1beta
Any "OpenAI-compatible" relay	OpenAI Compatible	The base_url from the relay's docs

OpenAI Compatible vs OpenAI Responses

• OpenAI Compatible (Chat Completions): the legacy endpoint that nearly every third-party relay supports. If you're not on the official OpenAI endpoint, you almost certainly want this one.
• OpenAI Responses: the newer endpoint OpenAI recommends for GPT-4.1+. It supports richer reasoning summaries, tool use, and persistent state across turns. Only the official OpenAI endpoint supports it.

When in doubt, pick OpenAI Compatible. Switch later if needed.

Full Walkthrough

Below uses OpenAI official as the example. Every other vendor follows the same flow — only the values you type in steps 4–6 change.

Step 1: Open the provider list

Chat → tap "Settings" in the bottom tab bar
→ under "Conversation" → "Providers & Models"
→ tap to enter

You see an empty (or near-empty) list titled Providers & Models.

Step 2: Add a new provider

Tap + Add Provider in the top-right (plus icon).

The Add Provider form opens, divided into four sections: Basic Info, Authentication, Proxy (per provider), Header Overrides.

Step 3: Fill in "Basic Info"

Field	What to put
Provider Name	Your display name. Shown in the model picker. e.g. "OpenAI Official", "My Relay", "Claude".
API URL	Paste the Base URL from the cheat sheet above. No trailing slash, and don't append /chat/completions.
API Format	Pick from the dropdown. OpenAI official defaults to "OpenAI Compatible".

Easy-to-miss Details

• No trailing slash (/v1, not /v1/)
• Don't forget https://
• If the relay docs give https://xxx.com/v1/, strip the trailing /

Step 4: Fill in "Authentication"

Field	What to put
API Key	Paste the key from the vendor console. Just the key, no Bearer prefix, no quotes.
Show Plaintext	Toggle. Turns on to reveal the key so you can double-check the paste. Turn it off after.

Multiple Keys per Provider

Want to rotate among several keys (to dodge per-key rate limits)? Concatenate them with English commas: sk-aaa,sk-bbb,sk-ccc. The app rotates through them in order on each request.

Step 5: Skip "Proxy (per provider)"

Leave Use Per-Provider Proxy (overrides global) off if your network can reach the vendor directly. Set up proxies after chat is working.

Step 6: Skip "Header Overrides"

Don't touch this on your first pass. It's for relays that demand custom HTTP headers (e.g. X-Org-ID).

Step 7: Save

Tap Save in the top-right. If anything's missing (Base URL empty, etc.), the Save button grays out and the form highlights what's wrong.

After saving you bounce back to the provider list with your new entry visible.

Step 8: Open the provider and add models

Tap the new provider row to enter Provider Details. The bottom of the page shows an empty Models section.

Two ways to add models:

1. Fetch from cloud — the cloud icon in the top-right (accessibility label: "Fetch from cloud"). Sends GET /models to the vendor and pulls back every model your key can use. Prefer this — saves time and avoids ID typos.
2. Add manually — the + button beside it (label: "Add Model"). Type the model ID (gpt-4o-mini) and an optional display name. Useful if the vendor has no /models endpoint.

Once models are listed, toggle on the row's switch. Models that aren't enabled won't show up in the chat picker.

Step 9: Run the connectivity test

There's a test button in the top-right of the provider details page (accessibility label: "Model Test"). It fires a minimal request with your key and reports back. Use it — you don't have to bounce back to chat just to verify.

Results are binary:

• ✅ Green "Test Passed" — good, move on.
• ❌ Red error — match the message against the table below.

Step 10: Set this model as "Current Model"

Back in Settings, the top Current Model section → tap Model → pick the model you just enabled.

Now you're truly done. Head to Start Your First Chat to send a message.

Common Errors

Error	Real Cause	Fix
401 Unauthorized / "Authentication failed"	Key is wrong / expired / has trailing whitespace	Re-copy from the vendor console; strip whitespace
404 Not Found	Base URL wrong, or path appended/missing	Re-check the cheat sheet
"Model not found" / model not found	Wrong model ID, or key doesn't have access	Use "Fetch from cloud" and pick from the list
"Connection timeout" / "Network error"	Can't reach the vendor / proxy needed	Set up the per-provider proxy below
429 Too Many Requests	Vendor rate limit triggered	Wait a few minutes; or upgrade your account quota

Advanced

Everything below assumes basic chat already works.

Custom Headers (Header Overrides)

Some relays or enterprise endpoints require extra HTTP headers, like X-Org-ID or non-standard Authorization formats.

Where: Edit Provider → Header Overrides → tap Add Expression.

Syntax: one entry per line, key=value:

User-Agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64)
X-Org-ID=org_abc123
Authorization=Bearer {api_key}

The {api_key} placeholder is substituted with the API key actually used for that request (if you set up key rotation, it becomes the rotated key). Useful for services that demand the key in a custom header instead of the standard Authorization.

Per-Provider Proxy

If you can't reach the vendor directly, configure a proxy for this provider:

Where: Edit Provider → Proxy (per provider) → toggle on Use Per-Provider Proxy (overrides global).

Field	Notes
Enable Proxy	Master switch for this proxy config
Proxy Type	HTTP / HTTPS or SOCKS5
Proxy Host	Hostname or IP, e.g. 127.0.0.1
Port	1–65535. Typical: HTTP proxy 8080 / 7890; SOCKS5 1080
Username (optional)	Proxy auth username; setting it enables auth
Password (optional)	Proxy auth password

Global vs Per-Provider Proxy

ETOS also has a Global Proxy (Settings → Providers & Models → Global Proxy). The per-provider proxy wins — when a provider's "Use Per-Provider Proxy" is on, the global proxy is ignored.

Useful when provider A needs a proxy but provider B works directly: only give A a proxy, leave B alone.

Multi-Key Rotation

The Authentication → API Key field accepts comma-separated keys:

sk-aaaaaaaaaa,sk-bbbbbbbbbb,sk-cccccccccc

Each request rotates through them in order. If a key trips 429 or auth fails, the next one is used.

Different Capabilities → Different Models

ETOS slots models by capability. A single provider can host GPT-4o as the chat model, whisper-1 as speech-to-text, and tts-1 as the TTS model without conflict.

Where: under Settings → "Conversation" group, Preferences / Text-to-Speech (TTS) / Speech Input / Image Generation / Daily Pulse each have their own "preferred model" picker.

Model-Level Advanced Parameters (Parameter Expressions / Raw JSON Body)

Inside Providers & Models, tap a specific model row to enter Model Settings. There you can:

• Override default sampling parameters (temperature, top_p, …) per model
• Write parameter expressions — a template syntax that computes parameter values dynamically (e.g. scale max_tokens with message length)
• Edit the raw JSON body sent in the request — last-mile customization for non-standard endpoints

Details in Chat & Models advanced section.

Next

Provider set up, connectivity test green, current model selected → Start Your First Chat