Solutions
Deploy your own finance agents in Slack. They code invoices, chase late payments and prepare the close. Nothing moves money without a person’s approval.
Payables
Coding overnight invoices
In beta alongside teams at
Perplexity
Warp
Corgi Cafe
FLORA
Composio
Palantir
Cognition
Every action that moves money or changes the books waits for the right person.
Each tool runs automatically, asks first, or stays off. Reads run on their own; writes ask first unless you change them.
The right person approves in Slack, and the agent carries on from there.
Choose who can view, use, edit or manage each agent.
Set daily, weekly or monthly limits and see exactly where spend goes.
Agents remember what your team teaches them, with clear limits on who can see what.
Our agreements with model providers prohibit training on your data. A DPA covers every customer.
Agents code invoices, chase payments and prepare reconciliations. Your team reviews and approves, and nothing moves money without a person.
Reconciliations, accruals and checklists are prepared before your team sits down to review.
Every overdue invoice gets a timely, personal reminder, and the largest ones reach you first.
Payments, refunds and journal entries wait for a person’s approval, in the thread.
Agents code invoices, chase payments and keep the ledger current.
Match invoices to purchase orders and code them for review.
Queue approved bills for each payment run and flag anything unusual.
Send reminders matched to each customer’s history, and escalate the large ones.
Check refunds against your policy and the account’s history before they are issued.
Check expenses against policy and ask for missing receipts.
Answer vendors’ payment status questions from the ledger.
Agents code, chase and reconcile overnight. Your team starts the day reviewing and approving, not entering data.
Twelve vendor invoices came in overnight. Ten match a purchase order and are coded in QuickBooks for review. Two don’t: AWS is $3,180 over its PO, and a Figma invoice has no PO.
The AWS overage is expected, approve it. Who owns Figma?
@Jack owns the Figma account. Jack, can you confirm this invoice?
Yes, that’s the annual renewal.
Thanks. Both are coded and queued for Thursday’s payment run, pending approval.
@Collections what’s overdue?
$184K is past due across 9 accounts, and four are over 30 days. Reminders are drafted to match each account’s history. The two largest are flagged for a call instead.
Send the reminders. I’ll call Northwind myself.
Sent from the AR inbox. The Northwind call is on your calendar at 2 PM, with the invoice history attached.
@Controller how close are we on month-end?
Bank reconciliations are done for all three accounts, and accruals are drafted for review. Two items are open: the payroll journal and the deferred revenue schedule.
@FP&A Analyst how is burn tracking against plan?
October burn was $412K against a $440K plan. Software is 12% over, mostly new AI seats. Headcount is two roles under plan. The variance breakdown is in the board model.
Each agent has one job, the tools for it and limits you set. Start with these, or describe your own.
Coding overnight invoices
Set agents up in the app in an afternoon, or create and manage them from your own systems.
Describe the job, connect your tools and add the agent to Slack. No engineering team required.
Create, configure and manage agents with a workspace API key, or let your coding assistant do it through the Oasis MCP server.
# Create an agent with a workspace API key curl -X POST \ https://api.joinoasis.com/api/v1/organizations/$ORG_ID/agents \ -H "X-API-Key: $OASIS_API_KEY" \ -d '{ "name": "Ledger", "agent_role": "Finance", "system_prompt": "Chase overdue invoices…" }'
Only if you allow it, and always with approval. Payments, refunds and journal entries ask a person first by default.
QuickBooks, Stripe, Google Sheets, DocuSign, Gmail and more. Agents can also use banking portals and web apps through their own computer.
Only the people you give access to. Access is set per agent, and memory has clear limits on who can see what.
No. Our agreements with model providers prohibit them from using the data we send them to train or improve their models.
Most teams have their first agent working in an afternoon: start from a template, connect your accounting system, and add it to Slack.
The same platform runs agents for every team in your company, each with its own tools and limits.
