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

# OpenClaw on SaladCloud Overview

> Understand what OpenClaw is, why model costs can grow quickly, and why SaladCloud is a strong hosting option.

*Last Updated: February 13, 2026*

OpenClaw is a local-first AI agent framework. It can chat, run tools, automate tasks, and connect to channels such as
Telegram, Discord, and other messaging platforms.

This page explains:

* what OpenClaw is,
* why model costs can grow quickly,
* why SaladCloud is a practical way to host models for OpenClaw.

## What OpenClaw is

OpenClaw is the orchestration layer around your model, tools, memory, and channels. The model is the "brain," while
OpenClaw handles:

* conversation and session state,
* tool execution and workflow logic,
* channel integrations,
* scheduling and background activity.

Because OpenClaw can run frequent background calls and tool-heavy workflows, your model choice and hosting strategy have
a direct impact on cost and performance.

## Why costs increase with API-hosted models

OpenClaw itself is free, but model inference is not. With traditional API providers, billing is typically token-based.
Costs tend to grow because of:

* accumulated conversation and tool context,
* repeated background/heartbeat interactions,
* sub-agent and multi-step task execution,
* always-on usage patterns.

In practice, this can make costs hard to predict for active agent workflows.

## Why host OpenClaw models on SaladCloud

SaladCloud changes the cost model from token billing to compute-time billing for your deployment. For many OpenClaw
workloads, this has three main advantages:

* **Predictability**: pay for runtime, not each generated token.
* **Control**: scale replicas up/down, including scheduled scale-to-zero windows.
* **Privacy posture**: keep more prompt/context flow within infrastructure you control.

For users running long sessions, heavy tool usage, or frequent background checks, this can materially reduce total cost.

## Inference options on SaladCloud

You can run OpenClaw-compatible model endpoints with:

* **Ollama**: easiest path for most users.
* **vLLM**: higher throughput and advanced serving optimizations.
* **TGI**: strong text-generation serving option for many Hugging Face models.

All can be used with OpenClaw when configured with compatible endpoints and model metadata.

## Recommended operating pattern

A common production pattern is:

1. run OpenClaw locally (or in your controlled environment),
2. host one or more model endpoints on SaladCloud,
3. set model fallbacks in OpenClaw,
4. apply autoscaling/scheduling to align runtime with active hours.

This balances capability, reliability, and cost.

## Next step

Follow the full setup guides to deploy OpenClaw with SaladCloud-hosted models:

* [OpenClaw + Ollama (Salad Hosted) + Telegram](/container-engine/how-to-guides/openclaw/openclaw-ollama-salad-hosted-telegram)

## References

* [Reduce Your OpenClaw LLM Costs: SaladCloud Guide](https://blog.salad.com/reduce-your-openclaw-llm-costs-saladcloud-guide/)
* [OpenClaw Documentation](https://docs.openclaw.ai/)
* [Ollama + OpenClaw](https://ollama.com/blog/openclaw)