Camunda-provided LLM
Run AI agents quickly in Camunda SaaS with Camunda-provided LLM.
About
Camunda-provided LLM is a Camunda-managed LLM provider option that comes with automatically configured credentials. With it, you can run AI agents in your processes right away without additional setup.
Camunda-provided LLM is only available in Camunda 8 SaaS. It is not available in Camunda 8 Self-Managed.
Camunda-provided LLM is free to use within the provided budget, and is intended for testing and experimentation. When you're ready for production or need more control, switch to a customer-managed provider.
Key benefits:
- No LLM account setup required. You don't need to sign up with a model provider or configure credentials to start exploring AI agents.
- Compare different models. Run your agent with different models and easily switch between them to find the best fit for your use case.
- No surprise bills. Your organization gets a free, preconfigured budget for testing and experimentation.
- Instant blueprints. AI agent blueprints that use Camunda-provided LLM work out of the box with no configuration needed.
- Seamless transition. When you're ready for production, switch to a customer-managed provider like AWS Bedrock without changing your process architecture.
Camunda-provided LLM is available in Camunda SaaS for:
- SaaS trial organizations: Includes Camunda-managed credentials and a free budget. Camunda-provided LLM is enabled by default.
- SaaS enterprise organizations: Includes a larger budget to support multiple proofs of concept. You must explicitly enable the Camunda Provided LLM toggle in Camunda Console. This is a separate toggle from AI-powered features, which controls other AI capabilities such as Camunda Copilot. The first time you enable either toggle, you must accept the Terms for AI Usage.
Availability, budgets, and UI may vary by environment and rollout stage.
See Trial vs. enterprise budgets for more details.
Set up Camunda-provided LLM
Once Camunda-provided LLM is available in your organization, its credentials are populated automatically as cluster secrets.
- If you are using an AI agent blueprint, no additional configuration is needed in most cases. Explore selected AI agent blueprints in the Camunda Marketplace.
- If you are building your own agent from scratch, enable Camunda-provided LLM by configuring your AI Agent connector with the following parameters:
- Provider:
OpenAI Compatible. - API endpoint:
{{secrets.CAMUNDA_PROVIDED_LLM_API_ENDPOINT}}. - API key:
{{secrets.CAMUNDA_PROVIDED_LLM_API_KEY}}. - Model: Select a model from the list of supported models. For example,
anthropic/claude-sonnet-4.6.
- Provider:
Supported models
Camunda-provided LLM uses a managed LLM gateway that supports the models listed below. You can switch between these models to compare how your agent performs with each one. When using the AI Agent connector, set the Model field to one of the following values:
| Model | Value to set in Model | What it's good for |
|---|---|---|
| Anthropic Claude Sonnet 4.6 | anthropic/claude-sonnet-4.6 | Best as the default for complex agent tasks, balancing strong reasoning, reliable tool use, speed, and budget consumption. |
| Anthropic Claude Haiku 4.5 | anthropic/claude-haiku-4.5 | Best for lightweight assistants, short interactions, and lower-cost tasks that still need good instruction following. |
When selecting a model, consider your process requirements, expected usage volume, and token budget. For model selection guidelines, see how to choose the right LLM.
Trial vs. enterprise budgets
The budgets, measured in dollars (USD) spent, differ depending on your SaaS plan:
- Trial: A smaller budget intended for quick evaluation and early experiments by individuals and small teams.
- Enterprise: A larger budget intended for broader team experimentation and proofs of concept.
Budgets are topped up automatically and enforced at the organization level (not per user). This means multiple users in the same organization draw from the same budget.
What the budget cover
The Camunda-provided LLM budget covers LLM provider calls during AI agent execution:
- Trial budget: Allows for a hundred to a few thousand agent runs, depending on the model used and the agent complexity.
- Enterprise budget: Is significantly larger to support more extensive experimentation.
Other Camunda AI features, such as Camunda Copilot, do not consume your Camunda-provided LLM budget and can be used independently.
The total cost of an agent run depends on how many LLM calls it makes, which can vary based on the agent’s design and task complexity. Cost also depends on the model used, since different models have different per-token pricing.
When budget is exhausted
When your organization reaches its Camunda-provided LLM budget cap:
- Additional LLM calls are blocked.
- Your process execution may fail with an “out of budget” error, such as
COST_LIMIT_EXCEEDED, depending on how your process handles errors.
If your process model doesn’t handle LLM failures, an exhausted budget may result in incidents or failed instances. Consider adding BPMN error handling to provide a user-friendly fallback path.
Monitor usage
The Camunda-provided LLM budget is shared across your organization, so you should monitor consumption. Camunda Console shows usage statistics for Camunda-provided LLM, including:
- How much of your budget has been used.
- How much budget remains.
Use this data to plan your transition to a customer-managed provider when you're ready for production.
Switch away from Camunda-provided LLM
As you move from evaluation to production, you may want to switch to your own LLM provider. This gives you:
- Direct control over provider choice.
- Your own billing and quota management.
- The ability to scale beyond the Camunda-provided LLM budget caps.
- Ensure your organization has access to the LLM provider you plan to use.
- Gather credentials and any required configuration.
- Identify where your current AI agent models rely on Camunda-provided LLM defaults.
To switch away, follow these steps:
- Add your LLM provider credentials in the appropriate Camunda location for managing secrets and credentials.
- Update your AI Agent connector configuration to use the new LLM provider.
- Re-deploy your process.
- Test a process instance end-to-end and verify results.
Your orchestration model doesn’t change during this transition. The BPMN process, event choreography, and human touchpoints you designed with Camunda-provided LLM carry forward unchanged, while only the LLM backend configuration shifts.
Data processing and AI terms
Review how Camunda-provided LLM processes your data, how OpenRouter is involved, and which terms apply when you use this feature.
How Camunda-provided LLM works
The Camunda-provided LLM is an optional feature. When enabled, you configure which process variables, prompts, and tool outputs the agent may use. Only the data you choose to make available to the agent is sent for processing. You control this through your process configuration, such as input mappings.
How OpenRouter is used
OpenRouter, Inc. is a routing service that gives Camunda access to a range of third-party AI models through a single integration. OpenRouter does not host or train models itself. Instead, it forwards each request to the selected AI model provider for processing. Camunda restricts this feature to providers that enforce zero data retention (ZDR). This means that neither OpenRouter nor the underlying model provider retains your data after processing the request, and your data is not used to train their models.
OpenRouter as a sub-processor
OpenRouter, Inc. is a sub-processor engaged by Camunda in connection with this feature. When you enable this feature, your prompts, agent memory, and tool call inputs and outputs, which may contain personal data, are transmitted to OpenRouter and the selected model provider for the purpose of generating a response. This feature is optional and can be disabled at any time by turning off the Camunda Provided LLM toggle in Camunda Console. When disabled, no data is sent to OpenRouter. A comprehensive list of Camunda's sub-processors is available in Camunda's Trust Center.
AI terms
This feature is an AI Feature under Camunda's Terms for AI Usage, which apply to your use of it unless your existing agreement with Camunda provides for AI terms, in which case the latter would prevail. In addition, please refer to our AI Usage Guidelines to learn more about how to use Camunda’s AI features responsibly.
Data sharing
You control what data is made available to this feature through your process configuration. Do not include sensitive data, or other content you are not authorized to share with a third-party AI service provider.