Transactional Finance Suite
Finance agentsfor manufacturers.
Vectorfield runs your AR, AP, and bank reconciliation. Your team approves what moves money. Audit-ready every day of the month.
- 01
- GDPR, day one
- 02
- EU region
- 03
- Hash-chained audit
- 04
- SOC 2 Type II in progress
The agents
Four agents. One finance manager.
The Transactional Finance Suite. The repetitive 60 to 80 percent of finance operations, automated. The rest stays with your team.
- 01
@collections
Replaces:AR clerk chase work
Sends the right reminder to the right customer at the right time. Knows who pays late, who is on a payment plan, who has an open dispute. Levels 1 and 2 autonomous in gated mode. Level 3 (formal demand) routed for approval.
Human gate:CFO if escalating to dispute
- 02
@invoice-intake
Replaces:AP clerk data entry and coding
Reads incoming bills from the AP inbox, extracts line items, codes to the chart of accounts, matches to POs and goods receipts, drafts the posting. Posts as draft to Business Central. Never pays without approval.
Human gate:Controller approves the coded draft
- 03
@reconciliation
Replaces:Bookkeeper bank matching
Matches bank transactions against open invoices and bills. Handles partial payments, short-pays, overpays, on-account receipts, bank charges, FX gains. Routes exceptions with evidence.
Human gate:Controller reviews exceptions
- 04
@reporting
Replaces:Light controller reporting
Answers cash-position, AR-ageing, AP-due, vendor-spend questions on demand. Pushes a Friday cash report autonomously. Read-only, no gate needed.
Human gate:None
@finance-manager orchestrates the four and routes anything that doesn’t fit to @human. The approval queue is the only path for actions that move money.
The trust spine
Three answers your CFO will ask for.
Every finance customer asks the same three questions about AI. The platform is built so each one has a concrete, audit-grade answer.
- 01
Is our data used to train your models?
No. Contractual in the MSA. Technical: every model call routes through a gateway. Restricted data classes hit zero-retention endpoints only, in the EU, with redaction policies the tenant controls. Reviewed quarterly per vendor.
Model gateway · per-tenant policy · per-vendor DPA
- 02
Can the AI move money or sign a filing without approval?
No. Money movement and statutory filings route to @human, keyed to your delegation-of-authority matrix. The runtime cannot execute a money-affecting action without an approval id in the call. Enforced at the tool gateway. Audited on every call.
Tool gateway · DoA matrix · approval queue
- 03
Can we prove what the AI did and why?
Yes. Every model call, tool invocation, policy citation, and inter-agent message is in a hash-chained audit log per tenant, exportable to your external auditor. Each agent decision cites the rule, the policy version, and the model version that produced it.
Hash-chained log · 7-year retention · auditor export
How an agent earns autonomy
Shadow. Co-pilot. Autonomous, with gates.
An agent moves up a mode only when it has earned it on your data. The numbers are explicit. The toggle is two-key. The demotion is one-key, yours.
- 00
Discovery Sprint
Two weeks at your premises. Interviews with AR, AP, bookkeeper, controller, CFO. Action-log mining. The finance operations manual, seed policies, golden eval cases. You keep the manual whether or not we ship.
Output: operations manual, ~20 eval cases per workflow
- 01
Shadow
The agent observes, proposes, never acts. Read-only tools only at the gateway. Your clerk grades the proposals daily. Disagreements become eval cases.
≥14 days · ≥50 graded proposals · ≥90% agreement
- 02
Co-pilot
The agent suggests in the Console. Your team accepts, edits, or rejects. Approved actions run with your team member as the audit actor. Edits become proposed preference rules.
≥28 days · ≥100 reviewed proposals · ≥85% acceptance
- 03
Autonomous, gated
The agent acts within policy. Anything over the DoA threshold routes to the approval queue. A P1 incident, an acceptance-rate drop, or a denied tool call demotes the agent automatically.
Two-key promotion · one-key demotion by your CFO
Promotion is per agent, per tenant, per workflow. None of the four MVP agents reaches full autonomy in the first 12 weeks. Reconciliation stays in co-pilot at end of MVP, deliberately.
The Console
One surface for the finance team.
Approvals you click through on a phone. An audit chain you can hand to your auditor. A cost meter per agent. A policy library your controller authors. The preference rules your team’s corrections produced this quarter. All scoped to your tenant.
- 01Approval queue
- 02Audit chain viewer
- 03Cost & token meter
- 04Policy library
- 05Preference rules
- 06Exception inbox
- 07Run-mode controls
- 08Connector status
tenant: AMT POSTPRO · 3 PENDING · €4,820
Approval queue
AP-2026-0418
Hellenic Power
Amount
€2,340.00
GL
6480
Cites
expense-policy v3
+ amt-vendor-hellenic-power-coding
From
@invoice-intake
Controller approval
AR-2026-1102
ACME Industries
Amount
€18,400.00
GL
(dispute)
Cites
dunning-policy v2
+ amt-customer-acme-soft-tone
From
@collections
CFO approval, level 3
REC-2026-0079
BNP Paribas
Amount
€480.20
GL
7150 (charges)
Cites
reconciliation-policy v1
+ amt-skip-below-5
From
@reconciliation
Controller exception
Posture
What the platform commits to.
The non-negotiables. Each item maps to a section of the platform documentation we share under NDA on request.
Multi-tenant from line one
Postgres row-level security. Tenant-scoped vector namespaces. Tenant-scoped message bus. Per-tenant encrypted secrets. No shared mutable state.
Hash-chained audit log
Append-only. Per-tenant chain head. Every money-affecting write, every inter-agent message, every approval, every policy citation. 7-year retention.
EU data residency
Supabase EU (Frankfurt). Cloudflare R2 EU. The model gateway denies non-EU endpoints for restricted data classes by default.
Model-agnostic
Every model call routes through a gateway. A better or cheaper model is a config change, not a migration. Open-source first for extraction; frontier for hard reasoning.
No fine-tuning on your data
Customization lives in skills, policies, preference rules, and RAG. The model is a stateless commodity. Contractual in the MSA.
Human-in-the-loop where it counts
Money movement and statutory filings route to the approval queue, keyed to your delegation-of-authority matrix. Enforced at the tool gateway.
GDPR from day one
DPA template, sub-processor list, data inventory, retention policy, DPIA template, per-tenant export and right-to-erasure. EU-controller, US-processor SCCs module 2.
SOC 2 Type II in progress
Continuous evidence collection from month 2. Type I audit month 7 to 8. Type II report month 13 to 14. Penetration test annually.
What this is not
Lines we don’t cross.
The opposite of a feature list. The clarity is the point.
payroll engine
We integrate ADP, Gusto, or Sage. We don’t run payroll.
tax engine
We partner with Avalara or Sovos. We assemble the filing, we don’t compute the tax.
treasury management
FX, hedging, letters of credit, and intercompany netting belong to Kyriba or GTreasury, not to us.
workflow builder
No drag-and-drop canvas. Agents collaborate through a message bus governed by tenant policy.
autonomous about money
An agent cannot move money or sign a filing. Ever. The approval queue is the only path.
replacement for your controller
Judgment stays with your team. Repetition moves to the agents.
Start with a Discovery Sprint
Two weeks. One finance team. One operations manual.
We spend two weeks at your office. We shadow your AR, AP, bookkeeper, controller, and CFO. You leave with a finance operations manual that is yours to keep, regardless of whether we ship.