> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tokios.com/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Tokios exposes private local models through https://api.tokios.com using an outbound-only connector tunnel.
> Use 'connector' for the local tokios-connector process.
> Use 'deployment' for the public model name in the API model field.
> Use 'upstream model id' for the model served by Ollama, llama.cpp, vLLM, LM Studio, or another local backend.
> API keys use the prefix sk-tok-. Connector tokens use the prefix ct-tok-.

# Tokios FAQ: Setup, Connectivity, and Troubleshooting

> Answers to common questions about Tokios setup, firewall requirements, API compatibility, streaming, data privacy, supported backends, and billing.

This page covers the most common questions and issues when setting up and using Tokios.

***

## Setup & connectivity

> [accordion] The connector starts but my requests return 503
>
> A `503` means the Tokios gateway can't reach an active connector tunnel for the requested model. Check that:
>
> 1. The connector process is still running and connected.
> 2. The model name in your API request matches a deployment you registered in the console.
> 3. The device was **approved** during pairing and shows as online in the console's [Connectors](https://tokios.com/console/connectors) tab. (If you run the advanced multi-model config-file path, also confirm the `ConnectorId` and upstreams in `tokios-connector.json` match the connector shown in the console.)

> [accordion] Do I need to open any firewall ports?
>
> No. The connector only dials outbound to Tokios over WebSockets (port 443). Your firewall needs no changes, and nothing on your machine listens for inbound connections.

> [accordion] Can I run the connector on a different machine than my model?
>
> Yes. Point the upstream's `BaseUrl` at the other machine — for example, `http://192.168.1.50:11434/v1`. No opt-in setting is needed: the connector forwards only to hosts named in its own config file, never to a URL supplied by the gateway. To lock an upstream to its host, add the host to `AllowedHosts`; a later `BaseUrl` edit that points somewhere else then fails at startup. See [Connector Config](https://docs.tokios.com/reference/connector-config.md).

***

## API compatibility

> [accordion] Which API surfaces does Tokios support?
>
> Three: OpenAI Chat Completions (`/v1/chat/completions`), Anthropic Messages (`/v1/messages`), and OpenAI Responses (`/v1/responses`), plus `/v1/models` and `/v1/messages/count_tokens`. The gateway translates Messages and Responses to Chat Completions for your model. Provider-specific features that depend on a vendor's own backend (OpenAI Assistants, Files, fine-tuning) are not available.

> [accordion] Can I use streaming responses?
>
> Yes. Set `"stream": true` in your request body. Tokios returns server-sent events in the standard OpenAI/Anthropic streaming format on all surfaces.

> [accordion] My client sends requests to /v1 — will it work?
>
> Yes. For SDK clients, set the base URL to `https://api.tokios.com/v1` (with the `/v1` suffix). For Anthropic-style clients that use `ANTHROPIC_BASE_URL`, set it to `https://api.tokios.com` (no `/v1` — the client appends the path). Check your client's docs for which form it expects.

***

## Local models

> [accordion] Does Tokios store my prompts or model responses?
>
> Not by default. Tokios routes each request through the tunnel to your local model and streams the response back. The metering behind your usage page records token counts, latency, status, and identifiers only, and response bodies are never stored.
>
> A workspace admin can turn on **request logging** in the dashboard's **Settings**. Tokios then keeps the request body of each API call for 7 days, encrypted at rest, with attachments stripped and secret-looking text redacted, and shows it under **Prompt** when you open a request in the request log. Only your workspace's users can read stored prompts: admins see all of them, members only those sent with their own keys, and someone you shared a key with only if you allowed it for that key. You can turn it off, or delete every stored prompt, at any time. See [Request logging](https://docs.tokios.com/setup/request-logging.md).

> [accordion] Which model backends are supported?
>
> Any server that exposes an OpenAI-compatible Chat Completions API: Ollama, llama.cpp, vLLM, and LM Studio are the most common. Set the upstream `BaseUrl` (including `/v1`) in `tokios-connector.json`. See [Backends](https://docs.tokios.com/setup/supported-backends.md).

> [accordion] Can I serve multiple models?
>
> Yes. A single connector can serve several models using `Routes` + `Upstreams`, and you register a deployment (public name) for each. You can also run one connector per model. See [Model Routing](https://docs.tokios.com/guides/model-routing.md).

***

## Account & billing

> [accordion] Is Tokios free?
>
> Yes — the community plan is free with no credit card required, giving you full access to run and evaluate Tokios with your own hardware.

> [accordion] How do I revoke an API key?
>
> In the console, go to **API Keys**, find the key, and click **Revoke**. The key stops authenticating immediately.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.