Getting started
What is SentientOne?
An AI platform for building agents on your own content and tools, shipping them through an API, widget or workflow, and tracing every request.
Concept2 min readUpdated 23 October 2026
SentientOne is an AI platform for building agents that answer from your own content, call your own systems and run in production. You set an agent up in the dashboard, then put it to work through a streaming REST API, a script tag on your website, or a step in a multi-step workflow. Every request leaves a trace, so you can see exactly what the agent retrieved, which tools it called and what it cost.
The parts
| Part | What it does |
|---|---|
| Agents | A system prompt, a model from OpenAI, Anthropic, Google Gemini or Groq on your own key, and settings such as temperature and output format. See Create an agent. |
| Knowledge | PDFs, crawled websites, snippets and Q&A pairs, searched by meaning and by keyword on every question, then reranked. See Knowledge overview. |
| Tools (MCP) | Your Model Context Protocol servers. The agent can call their tools in the middle of an answer. See MCP overview. |
| Workflows (beta) | Multi-step automations on a canvas: agents, Classify, Extract, guardrails, HTTP, Slack, email and webhooks. See Workflows overview. |
| Website widget | One script tag with a publishable pk-so- key that only talks to one agent from the websites you allow. See Install the widget. |
| REST API | POST /v1/chat/stream with your secret sk-so- key, answered as server-sent events. See API overview. |
| Tracing and analytics | Every request as an OpenTelemetry trace, kept for 90 days, plus charts per agent. See Tracing. |
| Credits | Prepaid. One credit per agent reply; knowledge costs credits when it's added. See How credits work. |
How one reply works
- Your app, the widget or a workflow sends a message to an agent.
- SentientOne checks the key, the agent's rate limit and your credit balance, and reserves 1 credit.
- If the agent has knowledge, the question runs a vector search and a full-text search over it, the two lists are fused and reranked, and the best passages go to the model with the question.
- The model answers on your provider key. If the agent has MCP tools, it can call them first, for up to 8 rounds.
- The answer streams back token by token as server-sent events, with the sources it used and every tool call.
- The request is recorded as a trace you can open in Tracing. If the reply failed, the credit is refunded.
Who it's for
Developers and product teams who want an agent in their product, on their website or inside their operations without building retrieval, tool calling, streaming, rate limiting, billing and tracing themselves. Typical uses are support agents grounded on a help centre, order tracking through your own API, internal knowledge search and email or ticket triage. See use cases.
What you need
- A SentientOne account. New accounts get 200 free credits, no card needed.
- A key from one model provider: OpenAI, Anthropic, Google Gemini or Groq. SentientOne doesn't supply model access; you pay the provider for tokens at their prices. See LLM keys.
- Some content for the agent to answer from, if it should answer about your business.
Next
Follow the quickstart to create an agent, add knowledge, test it and call it from your code.
Still stuck?
Email us with what you tried and any request id. A person replies.