Autonomous Agents

Autonomous agents are AI-powered processors that handle repetitive accounting tasks without human intervention. They run in the background, processing events as they occur -- turning emailed invoices into posted bills, categorizing bank transactions, matching payments to invoices, and proposing payment batches. You configure the agents, monitor their work, and intervene only when they encounter something they cannot handle on their own.

Built-In Agents#

The system includes ten specialized agents, each handling a specific domain. The bill processor receives vendor invoices via email, extracts data using AI, matches or creates vendors, and posts the bill for approval. The bank transaction processor categorizes incoming bank transactions, matches them to vendors and customers, and creates the appropriate GL entries. The reconciliation agent runs a multi-pass matching algorithm to connect payments with invoices, keeping your aging reports accurate.

The payment proposal agent analyzes outstanding payables and suggests payment batches based on due dates, discount opportunities, and cash availability. The asset creator identifies capital expenditure transactions and creates fixed asset records. The master data agent handles vendor and customer creation requests from other agents, checking for duplicates and validating data quality.

Agent Configuration#

Each agent has configurable settings that control its behavior. You can enable or disable agents individually, set concurrency limits (how many instances can run simultaneously), configure timeout durations, and adjust confidence thresholds that determine when an agent should act autonomously versus escalate to a human.

Agents declare which workflows they are authorized to execute and which tools they can access. This permission model ensures agents operate within defined boundaries -- the bill processor can create transactions and vendors, but not modify payroll data.

Event-Driven Processing#

Agents are triggered by events: an email arrives, a bank statement is imported, a transaction is posted, or a scheduled time is reached. The event queue manages the workload, ensuring events are processed in order with retry logic for transient failures. You can monitor the queue to see pending, processing, and completed events.

Agent Memory#

Agents maintain persistent memory that improves their performance over time. The bill processor remembers vendor mappings (this email sender corresponds to this vendor), learned categorization patterns, and processing baselines. This memory means the system gets more accurate and requires less human intervention as it processes more of your data.

Monitoring and Control#

You can view the status of all agents at any time -- how many are running, what they are processing, and how their event queues look. If an agent makes an incorrect decision, you can correct it and the agent learns from the feedback. Event statistics show processing volumes, success rates, and error patterns, giving you visibility into the health of your automated accounting pipeline.