Transaction Coding Agent
The Transaction Coding Agent codes bank and card transactions to the right general ledger accounts, VAT codes and cost centres using each client's history, supplier patterns and your coding rules. It handles the routine volume automatically and queues only uncertain items for the bookkeeper, each with a suggested answer and the reason behind it.
Ideal forBookkeepers
The difference
From 3-4 hours per client per month to 30-45 minutes
Your bookkeepers spend their month reviewing exceptions and advising clients instead of allocating bank lines one by one.
Done by hand
3-4 hours per client per month
With the Vanine agent
30-45 minutes
Time returned to your team
80%
Indicative figures. We measure your own baseline in the first fortnight so the numbers you see are yours.
Correct VAT codes from the start
Coding a transaction as standard-rated when it is exempt, or claiming input tax on entertainment, flows straight into the VAT201. The agent applies the right VAT code consistently so returns rest on clean data.
More clients per bookkeeper
Allocating hundreds of bank lines a month is the biggest time sink in a practice. The agent clears the repetitive volume so your bookkeepers handle more clients without overtime.
Consistency across the team
Different staff code the same supplier differently, which muddies management accounts. The agent applies one set of rules per client, every month.
Capabilities
Bank transactions coded correctly, so bookkeepers only review what truly needs judgement.
Built for bookkeepers, accounting practices and in-house finance teams that process bank and card transactions in Xero, Sage or QuickBooks for South African businesses.
History-based coding
Learns each client's past allocations by supplier, description and amount, and applies them to new bank and card lines.
VAT code assignment
Applies standard-rated, zero-rated, exempt and no-input-tax codes, flagging blocked items such as entertainment and passenger vehicles.
Cost centre and tracking
Assigns cost centres, branches, projects or tracking categories based on your rules and the transaction details.
Confidence scoring
Scores every suggestion and only queues low-confidence or unusual items, such as first-time suppliers or large once-off payments.
Suggested answers
Gives the bookkeeper a proposed account and VAT code with a short reason for each queued item, so review takes seconds.
Rule learning
Turns the bookkeeper's corrections into new client rules, so the queue gets shorter month after month.
How it works
From your systems to a result you sign off
The bookkeeper reviews and approves postings. Draft allocations are never finalised without approval.
Connect the places your bank transactions already live.
Bank feeds and statement imports from SA banks, card transactions and the client's chart of accounts and coding history in Xero, Sage or QuickBooks. Already use something else? We connect that too.
- OpenAI Frontier
- Your own AI platform or models
- Microsoft 365 (Teams, SharePoint, Outlook, Excel)
- Xero
- Sage Business Cloud Accounting and Sage Pastel
- QuickBooks Online
- SA bank feeds and statement imports
Runs headless on a schedule as new transactions arrive, with the review queue available in Teams or OpenAI Frontier.
Draft allocations in the accounting system and an exceptions queue with suggested answers for the bookkeeper.
Step by step
- 01
New bank and card transactions arrive through the feed or an import.
- 02
The agent matches each line to the client's history and rules.
- 03
It proposes accounts, VAT codes and cost centres with a confidence score.
- 04
Uncertain items go to the bookkeeper's queue with a suggested answer.
- 05
The bookkeeper reviews and approves the postings.
- 06
Corrections become new rules for next month.
Deployment
Runs inside the tools your team already uses
The same agent and the same cited output, delivered three ways: through the AI assistant you have rolled out, on a model you host yourself, or inside your own software.
In the frontier assistant you already pay for
We publish the agent as an MCP server. Add it as a connector and your team calls it from the chat they use every day, with no new tool to learn.
"Load the January bank feed"
- ledger
- Xero
- period
- January
On a model you host yourself
Run the agent on open-weight models in your data centre or private cloud. Your bank transactions never leave your network, which keeps POPIA and data residency straightforward.
- Llama
- Mistral
- Ollama
- or any OpenAI-compatible endpoint
Inside your own software
Trigger a run from your own workflows, and push the finished output into your ERP, GRC or reporting systems. We build the integration with your team.
- REST API
- Webhooks
- Batch jobs
- Embeddable review panel
Governed the same way, every route
Whichever way your team reaches the agent, identity, permissions and audit stay with your existing controls.
- Single sign-on through Entra ID, Okta or Google Workspace
- Reads only the documents each user is already allowed to open
- Every tool call written to an audit log you can export
How we deliver it
One agent, three ways to hold it
Start where your team already works, then take it as far into your systems as the value justifies.
DATA SECURITY & PRIVACY
Your bank transactions stay yours.
The agent runs inside your tenancy, reads only what each task needs, and never trains on your bank transactions. Every step is logged so your auditors can see exactly what it touched.
Built for regulated work.
+YOUR BANK TRANSACTIONS NEVER LEAVE YOUR ENVIRONMENT
More agents
More agents we build for South African enterprises
Each one is built on the same foundation: your documents, your rules, and your people signing off.
Precision AI for Institutional Workflows
