> ## Documentation Index
> Fetch the complete documentation index at: https://hastekit.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Model Configuration
> Pick a provider, pick a model, and tune sampling and reasoning parameters
Every agent starts with a model. HasteKit talks to eight providers out of the box and gives you full control over the generation parameters each one accepts — including the reasoning controls on the newer reasoning-capable models.
## Supported providers
| Provider | Notes |
| --------------- | --------------------------------------------------------- |
| **OpenAI** | GPT-5, GPT-4.1, o-series, image and audio models |
| **Anthropic** | Claude Opus, Sonnet, Haiku, Fable |
| **Gemini** | Gemini 3.1 Pro / Flash family |
| **xAI** | Grok models including `grok-imagine` for image generation |
| **AWS Bedrock** | Bedrock-hosted Claude, Llama, Titan, and others |
| **OpenRouter** | Aggregator endpoint to 100+ models |
| **Ollama** | Local models served over an Ollama-compatible endpoint |
| **ElevenLabs** | Speech synthesis and transcription models |
Providers and their models are configured at the organisation-level under **Providers**. Once a model is registered there, it shows up in the **Model ID** dropdown here.
## Required fields
* **Provider** — which provider account to call.
* **Model ID** — the specific model from that provider.
## Generation parameters
All parameters are optional. Leaving a field empty uses the provider's default.
| Parameter | Range | Effect |
| ----------------------------- | ----------------------------- | ----------------------------------------------------------------- |
| **Temperature** | 0.0 – 2.0 | Randomness. Higher = more creative, lower = more deterministic. |
| **Top P** | 0.0 – 1.0 | Nucleus sampling. Lower values restrict the token distribution. |
| **Max Output Tokens** | integer | Hard cap on the response length. |
| **Max Tool Calls** | integer | Cap on tool calls per single LLM response. |
| **Top Logprobs** | integer | Return the top-K most likely tokens with their log probabilities. |
| **Parallel Tool Calls** | toggle | Allow the model to emit several tool calls in one turn. |
| **Reasoning Effort** | Default / Low / Medium / High | For reasoning models — how hard to think before answering. |
| **Reasoning Budget (Tokens)** | integer | For reasoning models — max tokens spent on internal reasoning. |
## Provider-specific extras
Some providers accept extra fields that don't fit the common set above — e.g. Anthropic's `cache_control` for [prompt caching](/docs/gateway/llm/sdk-integrations/anthropic) or `thinking` for extended-thinking. The Model tab exposes a JSON extras field where you can drop these in verbatim; they're passed through unchanged.
## Iteration cap
The **General** tab also exposes **Max Iteration** — the maximum number of agent loop turns (LLM call → tool calls → LLM call → ...) before the run is force-terminated. Leave empty to run without a limit. Use this as a safety net for agents that might otherwise loop indefinitely.