Agentic workflows, new agent architecture, skills, and more

Overview

New features:

Workflows: deploy AI agents as multi-step workflows that run on a schedule, analyze your data, and suggest actions you can approve directly
Agent actions in chat: the AI Agent can now draft payments, sweeping rules, loans, categorization rules, and more directly in conversation for you to review and approve
Treasury Policy: upload your treasury policy, let AI parse it into structured rules, and monitor compliance across your organization
Skills: create reusable instruction sets that shape how the AI Agent responds in any conversation

Improvements:

Approve all: approve multiple AI-suggested actions at once
Inline editing: edit AI-drafted resources before confirming them
Agent permissions: new permission scopes for agents in Access Group settings
Conversation history preserved up to 500 messages

Workflows

Deploy AI agents as recurring, multi-step workflows. Each workflow runs on a schedule you define, analyzes your financial data, and delivers structured output broken down by step.

Configure workflows from a catalog of agent templates, including Cash Position, Payments Brief, Forecast Variance, FX Risk Assessment, and more. Each template provides a suggested instruction, role, and set of analysis steps that you can customize. Assign an IAM role to the workflow so it only accesses the data it should.

When a workflow step is configured to suggest actions, the AI Agent can propose credit transfers based on its analysis. Suggestions appear in a table where you can review each one, open it to edit amounts or recipients, then approve or reject individually or in bulk.

Track every run in the Runs tab. Browse past executions, see per-step status, and trigger a manual run at any time with the Run now button.

Agent actions in chat

The AI Agent can now create and update resources directly in conversation. When the agent identifies an action to take, it presents an intent card you can review, edit, and approve without leaving the chat.

Supported actions:

Create credit transfers, with a full inline form for editing amounts, recipients, and payment details before approval
Create sweeping rules, using the same form as the rest of the product
Create loans and facilities for inter-company debt tracking
Create and update categorization rules

Each suggested action has an approve or reject action. If the agent suggests multiple actions in one response, use Approve all to confirm them in one click. After approval, a link takes you directly to the created resource.

The agent adapts its suggestions to what you are working on. If you are looking at a specific account or payment, the agent uses that context to make more relevant proposals.

Treasury Policy

Upload your treasury policy as a text or markdown file, or paste it directly. The AI parses your document into individual policy rules, each tagged with a domain: Banking, Debt, FX Risk, Governance, Investment, Liquidity, or Payment.

Parsed rules land in a review queue. Edit the rule text, assign a scope (global, entity, account group, or specific account), and mark each rule as "Looks good" to activate it or "Needs work" to keep refining.

Once active, enable compliance monitoring per rule. Monitored policies show a live compliance status - Compliant, In breach, or Unable to monitor - so your team always knows where you stand relative to policy.

Treasury Policy is available under Compliance in the main navigation.

Skills

Create reusable Skills that shape how the AI Agent responds. A Skill is a named instruction set you write once and attach to any conversation. For example, "Always report in EUR and focus on liquidity risk" or "Summarize balances by entity, not by account."

Manage Skills from the Skills page under Agents: create, edit, and delete them. In any conversation, attach one or more Skills from the picker before sending your message. The agent uses your attached Skills as additional context, adapting its analysis and tone accordingly.

Skills are stored centrally, so every team member with the right permissions can use the same playbook.

Payment and transaction analysis

The AI Agent now understands your payments and transactions, expanding beyond the initial release that focused on balances and cash flows.

Ask questions like:

Show me all failed credit transfers this week
Why was this payment rejected?
What is our payment volume trend over the past three months?

The agent can search credit transfers, direct debits, and booked transactions. It can look up individual payment details, investigate why a payment failed, and analyze payment volumes and patterns. Combined with the existing balance and cash flow analysis, the agent now covers the core data you work with daily.

The agent also understands what page you are on. When you open a transaction or credit transfer detail page and click the insights button, the agent starts a conversation with that context already loaded.

Additional improvements

Approve all: When the AI Agent suggests multiple actions in a workflow or conversation, approve them all at once instead of one by one. If any suggestion has a validation issue, the failing card expands so you can fix it.
Inline editing on approve: Edit AI-drafted credit transfers, loans, facilities, and sweeping rules before approving them. The agent prefills the form; you adjust and confirm.
Agent deletion cleanup: When you delete an agent, linked notification rules and workflows are automatically removed.
Agent permissions: New permission scopes for agents are available in Access Group settings, so admins can control who can create, view, edit, and delete agents.
Analytics view creation: Ask the AI Agent to build analytics views for you. Describe what you want to see and the agent creates a saved view you can revisit anytime.
Conversation history: Chat history is now preserved up to 500 messages per conversation, so longer analytical sessions are fully retained.
Ask Atlar shortcut: Open the AI Agent from anywhere in the dashboard with Cmd+K (or Ctrl+K).