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

# Providers Overview

> Connect Pioneer to cloud, local, OpenAI-compatible, and CLI-backed model sources.

Pioneer needs a model source before it can answer in threads or run model-backed tasks. That source can be a cloud API, a local model runtime, or an OpenAI-compatible endpoint.

Some agent tools, such as Codex CLI, are configured separately from ordinary API providers because they run through their own CLI workflow on the gateway host.

Provider setup follows the same flow for each provider. Connect the desktop app to the gateway, open **Providers** from the bottom bar, choose a provider type, enter its credentials or endpoint details, save it, and select a model in a thread.

Provider configuration belongs to the current workspace on the gateway, not to the desktop app. The gateway's secret store holds API keys and tokens. If you configure a provider in `Default Workspace` and switch to `Client A`, configure that provider again for `Client A`. A different gateway has its own provider configuration.

<Note>
  Provider requests are made by the gateway host. A remote gateway needs network access to the provider even if your desktop computer already has access.
</Note>

## General connection flow

Use this flow for any supported provider:

<Steps>
  <Step title="Choose the gateway and workspace">
    Connect the desktop app to the gateway and switch to the workspace that should own this provider configuration.
  </Step>

  <Step title="Open Providers">
    Click the providers icon in the bottom bar to open **Providers**.
  </Step>

  <Step title="Add the provider">
    Choose the provider type and enter the required key, endpoint, region, deployment, or local runtime address.
  </Step>

  <Step title="Save in the workspace">
    Save the provider configuration and confirm it appears for the current workspace.
  </Step>

  <Step title="Test with a thread">
    Create a thread, select the provider and model, and send a short test message.
  </Step>
</Steps>

Some providers can list models automatically. Others require you to enter the exact model ID. If model listing does not work, copy the model ID from the provider dashboard or runtime and test it with a small prompt.

When a listed model exposes reasoning controls, Pioneer can show a reasoning effort selector next to the model selection. Changing effort affects the next turn that uses that model selection. If the model does not advertise reasoning effort options, the selector is hidden and Pioneer sends the turn without an explicit effort value.

## Supported providers

Pioneer supports dedicated provider integrations, local runtimes, OpenAI-compatible endpoints, and separate CLI agent runtimes.

| Provider type                      | Supported providers                                                                                                                                                                                                                                                                                                               |
| ---------------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Direct cloud providers             | Anthropic, OpenAI, Google Gemini, OpenRouter, Azure OpenAI, AWS Bedrock, Copilot, GLM, Telnyx                                                                                                                                                                                                                                     |
| Local runtimes                     | Ollama                                                                                                                                                                                                                                                                                                                            |
| Built-in local services            | `Local` for gateway-local voice transcription and memory embeddings; it is not a chat provider                                                                                                                                                                                                                                    |
| OpenAI-compatible hosted providers | Groq, Mistral, xAI, DeepSeek, Together, Fireworks, Novita, Perplexity, Cohere, Venice, Cerebras, SambaNova, Hyperbolic, DeepInfra, Hugging Face, AI21, Reka, Baseten, Nscale, Anyscale, Nebius, Friendli, Lepton, SiliconFlow, AIHubMix, Astrai, StepFun, Baichuan, Yi, Hunyuan, OVHcloud, NVIDIA NIM, Synthetic, Doubao, Qianfan |
| OpenAI-compatible local servers    | LM Studio, llama.cpp, SGLang, vLLM, Osaurus, LiteLLM                                                                                                                                                                                                                                                                              |
| CLI-backed model sources           | Codex CLI and other supported CLI workflows installed on the gateway host                                                                                                                                                                                                                                                         |

Use [Supported Providers](/providers/others) when you need the broader compatibility list. Use [CLI Runtime API](/protocol/cli-runtime) only when you are building or debugging CLI-backed model selection.

## Choosing a provider

Start with a provider you already have access to. Use your organization's cloud account when it manages one, a local runtime when the gateway host provides it, or a router when you want several models behind one integration.

Model sources differ in tool support and operational behavior. Before using one for tool-heavy or scheduled work, confirm access, supported tool features, authentication, context size, cost, and latency.

For models with reasoning controls, start with the provider's default effort unless you know the tradeoff you want. Higher effort can improve harder reasoning tasks, but it may also increase latency or cost depending on the provider.

<AccordionGroup>
  <Accordion title="The provider does not appear in a thread">
    Make sure you configured it in the same workspace and gateway you are currently using.
  </Accordion>

  <Accordion title="Model listing fails">
    Try entering a model ID manually. Some providers do not expose model listing in a way Pioneer can use.
  </Accordion>

  <Accordion title="Authentication fails">
    Check the key, token scope, account status, endpoint, region, deployment name, and gateway environment.
  </Accordion>

  <Accordion title="A remote gateway cannot call the provider">
    Check network access, proxy settings, firewall rules, and DNS from the gateway machine.
  </Accordion>
</AccordionGroup>

## Further reading

* [Providers Architecture](/architecture/providers) explains provider adapters, model discovery, streaming, and attachments.
* [Secret Storage](/architecture/secrets) explains where provider keys are stored.
* [Gateway Architecture](/architecture/gateway) explains why provider calls happen from the gateway host.