Large language models and AI services in autonomous agents
Learn about the large language models (LLMs) and AI services that Reltio autonomous agents uses, the capabilities they support, and the option to use your organization's own LLM.
Autonomous agents use large language models (LLMs) to understand requests, plan tasks, interact with Reltio tools, and generate responses. By default, autonomous agents uses Anthropic Claude models through Amazon Bedrock.
Autonomous agents also supports Bring Your Own LLM (BYO LLM), which allows organizations to use their own approved LLM providers for supported agent capabilities.
Default LLMs and AI services
Autonomous agents use the following models and AI services by default.
| Model or service | Developer | Hosted on | Used for |
|---|---|---|---|
| Claude Sonnet 4.6 | Anthropic | Amazon Bedrock | Agent conversations, long-running tasks, specialized agents, reports, summaries, custom instructions, and custom agents. |
| Claude Haiku 4.5 | Anthropic | Amazon Bedrock | Guardrails that check prompts and agent responses against the autonomous agents usage policy, and system prompt security scans in custom agents. |
| Gemini 2.5 Flash | Google Cloud Vertex AI | Web search and webpage reading when web search is enabled for an agent. | |
| Parallel | Parallel Web Systems | Parallel web search service | Web search and webpage reading when web search is enabled for an agent. |
Web search is disabled by default. When enabled, autonomous agents uses either Google Cloud Vertex AI or Parallel to process web search requests.
Reltio periodically updates its default models to newer versions.
How autonomous agents uses LLMs
Autonomous agents use different LLMs and AI services for agent operations, task execution, agent creation, security checks, and web search.
The following table describes how each capability uses an LLM or AI service and whether it supports BYO LLM.
| Capability | What the LLM does | Default model or service | Supports BYO LLM |
|---|---|---|---|
| Agent conversations | Understands requests, determines which tools to call, and generates responses. | Claude Sonnet 4.6 | Yes |
| Long-running tasks | Plans and executes tasks using Reltio tools and generates task reports. | Claude Sonnet 4.6 | Yes |
| Long-running task eligibility check | Determines whether a request is suitable for execution as a long-running task. | Claude Sonnet 4.6 | No |
| Specialized agents | Performs multi-step workflows such as record analysis, address verification, web research, scoring, and data correction. | Claude Sonnet 4.6 | Yes, except for certain capabilities, such as guardrails and long-running task eligibility checks. |
| Custom instructions | Determines which custom instructions apply to a request and follows them. | Claude Sonnet 4.6 | Yes |
| Custom agents | Generates agent designs from descriptions and drafts supervisor prompts. | Claude Sonnet 4.6 | No |
| Guardrails and system prompt security scan | Checks prompts, responses, and agent system prompts against the autonomous agents usage policy. | Claude Haiku 4.5 | No |
| Web search | Searches the web and reads webpages when web search is enabled. | Gemini 2.5 Flash or Parallel | No |
Models autonomous agents does not use by default
Autonomous agents do not use the following models or model categories as part of its default configuration:
- Self-hosted and open-weight models: Autonomous agents does not self-host or run open-weight models as part of its default configuration.
-
Models from other providers:
Autonomous agents
does not use models from the following providers by default:
- DeepSeek
- Qwen (Alibaba)
- Kimi (Moonshot AI)
- GLM (Zhipu AI)
- MiniMax
- ERNIE (Baidu)
- Hunyuan (Tencent)
- Doubao (ByteDance)
- OpenAI
- Meta (Llama)
- Mistral
- Cohere
- Amazon (Nova or Titan)
These exclusions apply to Reltio-managed models. Organizations using Bring Your Own LLM (BYO LLM) can select models available through their configured providers and endpoints.
Bring Your Own LLM
Bring Your Own LLM (BYO LLM) allows your organization to use its own LLM endpoints instead of Reltio-managed models for supported autonomous agents capabilities.
BYO LLM supports agent conversations, long-running tasks, custom instructions, and actions in specialized agents. Certain capabilities, including guardrails, custom agents, long-running task eligibility checks, and web search, continue to use Reltio-managed models or services.
Autonomous agents support the following BYO LLM providers and authentication methods.
| Provider | Supported authentication |
|---|---|
| Amazon Bedrock in your organization's AWS account | IAM role, or access key and secret key |
| Anthropic | API key |
| OpenAI-compatible endpoints, such as OpenAI or your organization's LLM gateway | API key, OAuth, or OAuth with mutual TLS |
Reltio configures BYO LLM for your tenant through a support request.