Agents complete steps inside your processes. Assistants help your teams through their cases. Every action runs under the permissions you set, and lands in the same audit trail as a person’s.
Three roles, each with its own controls, pointed at the most complex, most regulated work you run, where every action has to be permissioned, explainable and recorded.
Placed on a process step or inside a user flow, an agent analyses information, produces an output, makes a decision or takes action, within the permissions you grant it.
Sitting inside the task, an assistant summarises documents, answers questions about the case and drafts content, while the person keeps the decision.
Connected through MCP, your AI agent builds or changes an application in Aurachain, reviewed by your team before anything reaches production.
On a process step, inside a user flow, or on the document the process produces.
Your teams stop doing the reading and keep doing the deciding.
Errors are caught in the user flow where they happen, not three steps downstream.
No rule to write for every document variation, and nothing re-typed from the case.
A chat window inside the task, working with that task’s data and documents.
Task owners spend their time on judgement rather than on reading.
Process agents, user flow agents and document agents are all defined in the AI Agent Builder. Configure one once and reuse it across the application, built by the people who own the process, using plain instructions, no data science team required.
Intent AI is the connection between Aurachain and your own AI agent. Through MCP, your agent builds a complete application, the process, the screens, the data model, the connections to your core systems, the documents and the agents inside it, or changes one already running, including the related changes needed to keep it working.
The agent can access only what the person who connected it is permitted to access. Your team reviews what it produced before anything reaches production.
Your agent does the building. Aurachain keeps the rules.
See the full build surface
Some requirements need a script. Describe what the application should do in plain language and the assistant turns it into working code, with in-context guidance on Aurachain’s custom code capabilities and refinement of what it generates.
The difficult parts of an application get built in the same place as the simple ones.
You choose which steps an agent completes, which steps a person keeps, and where an assistant helps instead. Most processes run as a mix, and an assistant helping a person is as common as an agent completing a step.
The line between agent and person is a setting you own, not a default you inherit.
An agent acts only when and where you instruct it to.
When the regulator asks who did this, the answer is the same shape whether it was a person or an agent.
How governance worksConnect agents and assistants to Anthropic, OpenAI or Azure OpenAI models within your organisation’s approved environment. Configure each agent to use the model best suited to its purpose, and switch the model behind an agent without redesigning the process or application around it.
Three domains where these design decisions show up as evidence, not just uptime.
Client lifecycle compliance, onboarding and ongoing due diligence, watch list and adverse media review.
Learn about compliance solutionsThird-party risk assessments, controls testing, and continuity requirements on the systems running regulated processes.
Learn about risk solutionsFraud investigation and case management, with alert triage and disposition recorded as it happens.
Learn about fraud solutionsAurachain is the execution layer where regulated institutions turn obligations, risk signals, and fraud alerts into governed, audited action.
An AI agent is configured for a specific task and can work autonomously as part of a process or user flow. An AI assistant is an agent that users interact with directly through a chat interface, helping them ask questions, review information, generate content or complete the work in front of them.
It can read and classify a document, check a name against a list, validate a submission, pull data from a connected system, draft a response, or make a decision within its permissions. It works only with the information and actions made available to it within the process.
Only the data and context you give it access to, within the permissions of the process it works in.
No. You choose which steps an agent takes and which stay with your teams, step by step. Most processes run as a mix, and an assistant helping a person is as common as an agent completing a step.
You can set a confidence threshold that determines when the agent hands the work to a person. You can also configure another agent to review its output and either approve it or send it back for correction. Every step and decision is recorded, so your team can see what happened.
Yes. Every action an agent takes is written to the same audit trail as a person's, with the same detail.
Yes. Put an assistant in one screen, or give an agent one step, and expand its role as trust builds. You can change where and how it works whenever you need to.
Intent AI connects your own AI agent to Aurachain through MCP. It builds the process, the screens, the data model, the functions, the documents and the agents inside the application, or changes an application you already run. Your people review the result before anything is deployed.
No. Intent AI assembles applications and changes them, and your people review everything before it is deployed.
As many as you need, across as many applications as you need. There is no limit on how many run at the same time.
Whichever you choose. The platform is model agnostic, so you can change providers without rebuilding the process.
No. Agents are built in the AI Agent Builder by the people who own the process, using plain instructions and a template.
AI is one layer of it. Here’s what it sits inside.
Build the operating layer for compliance, risk and fraud. Orchestrate what matters. Protect what counts.
Talk to Aurachain