The autopilots

Eight autopilots. One company.

Every back-office job at FormBridge is owned by a dedicated AI agent built against the same contract — a defined outcome, its own tools, reproducible decisions, and a published cost budget. Not chatbots. Not copilots. Autopilots.

Sample data — illustrative agent metrics

Formation

84 runs · 30d

Success: 97%
p95 cost: $0.34

Last run · 7m ago

EIN

79 runs · 30d

Success: 94%
p95 cost: $0.12

Last run · 22m ago

Banking

68 runs · 30d

Success: 92%
p95 cost: $0.41

Last run · 48m ago

Payments

52 runs · 30d

Success: 89%
p95 cost: $0.46

Last run · 2h ago

Bookkeeping

412 runs · 30d

Success: 99%
p95 cost: $0.08

Last run · 3m ago

Tax

41 runs · 30d

Success: 96%
p95 cost: $2.84

Last run · 2h ago

Compliance

187 runs · 30d

Success: 98%
p95 cost: $0.18

Last run · 11m ago

Co-pilot

924 runs · 30d

Success: 99%
p95 cost: $0.04

Last run · 1m ago

The autopilot contract

Six rules every agent must pass.

If a proposed agent can't meet all six, it doesn't ship. This is how we keep the product honest.

Rule 1
Defined outcome
Every agent names the artifact it produces — a filed Articles of Organization, a reconciled P&L, a signed 1120. If there's no clear outcome, there's no agent.
Rule 2
Owns its tools
Each agent calls registered-agent APIs, IRS submission, Plaid, banking portals directly through MCP tool servers. Not a chatbot handing work off to a human team.
Rule 3
End-to-end on the happy path
No user nudging required once the agent starts. You kick it off; it finishes. Interruptions only happen for real inputs the agent genuinely needs.
Rule 4
Escalates on uncertainty
Below the published confidence threshold, or on unsupported jurisdictions, work routes to an ops operator. No improvising on regulated matters.
Rule 5
Reproducible decisions
Every LLM call is logged as an AgentRun: prompt, tools called, output, cost, latency, outcome. Customer-exportable. Ops-replayable.
Rule 6
Compounds over time
Prompts, evals, and cassettes live in version control. Every regression is caught before deploy. Models get cheaper and better — that compounds directly into margin.

The lineup

Every agent, with receipts.

Cost budget is what we assert in CI — if an agent runs above the p95 target, the test fails. Escalation rules and out-of-scope lines are published here so you know exactly what you're paying for and what you're not.

Agent 01

Formation

Entity + jurisdiction + filing

Cost budget (p95)

$3.00 per run (p95)

Model

Claude Sonnet 4.6 — Opus for complex multi-jurisdiction recommendations

Outcome

A filed legal entity in your chosen jurisdiction, with Articles and Operating Agreement in your vault.

The Formation Agent interviews you in the onboarding wizard, recommends the right entity and jurisdiction with tradeoffs cited from the jurisdiction rules engine, drafts your Articles and Operating Agreement, and files them via a registered-agent partner API. It tracks state acknowledgement and hands off to the EIN Agent when filing completes.

Tools

Registered-agent filing API (Northwest / Harvard / partner)
Document template engine (WY LLC, DE C-Corp, UK Ltd, UAE FZ)
Jurisdiction rules engine (docs/PLAYBOOKS.md)
Document vault (OA, Articles — filesystem/S3)

Escalation

Unsupported jurisdictions, low-confidence recommendations (< 0.85), and non-standard cap tables escalate to a human operator before filing.

Out of scope

Legal advice. Immigration/visa work. Cross-border structuring beyond US/UK/UAE. Tax optimization across multiple entities.

Full Formation Agent page →

Agent 02

EIN

Federal tax ID on autopilot

Cost budget (p95)

$1.00 per run (p95)

Model

Claude Sonnet 4.6

Outcome

An IRS-assigned EIN in your document vault with the CP 575 confirmation letter.

The EIN Agent prepares Form SS-4 from your Formation Agent data, submits through the appropriate pathway (online for US-resident founders, authorized third-party designee for non-residents), and polls until the EIN is assigned. It uploads the IRS CP 575 letter to the vault and updates the company record.

Tools

IRS SS-4 online submission
Third-party designee pathway (non-residents)
IRS status polling
Document vault upload

Escalation

Rejected forms, IRS delays beyond 21 days, and applications flagged for additional review route to ops.

Out of scope

ITIN applications (separate workflow). State-level tax IDs (handled by Compliance Agent). Responsible-party changes.

Full EIN Agent page →

Agent 03

Banking

Application prep + anti-rejection

Cost budget (p95)

$2.00 per run (p95)

Model

Claude Sonnet 4.6

Outcome

A complete bank application pack assembled from your company data and submitted to Mercury, Wise, Relay, or Brex — the bank opens the account.

The Banking Agent fills Mercury, Wise, Relay, Brex, Revolut, and WIO applications from your company data. It coaches you on the exact KYC documents each bank expects, flags rejection risks (restricted industries, founder country, business activity mismatch), drafts your source-of-funds narrative, and submits the pack to the provider with the best approval odds. The bank — not FormBridge — decides and opens the account.

Tools

Per-bank application schema + requirements
KYC document checklist generator
Restricted-industry scanner
Source-of-funds narrative drafter

Escalation

Complex KYC cases, customers in bank-restricted geographies, and post-submit follow-up route to a partner banking specialist.

Out of scope

Opening or holding the bank account itself (the bank decides; FormBridge delivers a complete application pack). Fronting capital. Correspondent banking. Non-supported banks. Rate negotiation.

Full Banking Agent page →

Agent 04

Payments

Stripe / PayPal / Square approval

Cost budget (p95)

$2.50 per run (p95)

Model

Claude Sonnet 4.6

Outcome

A payments-processor application that passes KYC the first time.

The Payments Agent audits your site copy, business description, MCC code, and source-of-funds story against each processor's published policies before you submit. It rewrites vague descriptions, flags restricted-industry triggers, and drafts the exact application text that maximizes approval odds for Stripe, PayPal, or Square.

Tools

Stripe restricted-businesses policy scanner
Website content analyzer (About/Terms/Privacy checker)
Business-description rewriter
MCC code recommender

Escalation

High-risk industries, prior rejections, and post-rejection appeals route to ops.

Out of scope

Direct processor negotiations. High-risk merchant account shopping. Chargeback dispute drafting (roadmap).

Full Payments Agent page →

Agent 05

Bookkeeping

Monthly close, auto-run

Cost budget (p95)

$0.10 per 100 transactions (p95)

Model

Claude Haiku 4.6 — Sonnet for narrative synthesis

Outcome

Monthly P&L, balance sheet, and anomaly report delivered by the 5th of the month.

The Bookkeeping Agent ingests transactions from Plaid / Mercury / Wise feeds, auto-categorizes every transaction with a confidence score, routes anything below 95% confidence to human review, reconciles monthly, and produces P&L, balance sheet, and a narrative report. Every categorization decision is logged to an immutable event ledger.

Tools

Plaid / Mercury / Wise feed ingestion (scheduled sync)
Chart-of-accounts generator (per entity type)
Categorization classifier + confidence scorer
Reconciliation engine
Monthly narrative writer

Escalation

<95% confidence categorizations, reconciliation breaks, and anomalies above a configurable threshold route to human review.

Out of scope

Multi-currency consolidation (roadmap). Transfer pricing. Audit defense. Inventory accounting for e-commerce (roadmap).

Full Bookkeeping Agent page →

Agent 06

Tax

Returns drafted, CPA signed

Cost budget (p95)

$15.00 per return (p95)

Model

Claude Opus 4.7

Outcome

Tax returns drafted for your jurisdiction and entity (1120, 5472, 1120-S, 1040-NR, UK CT600), reviewed, signed, and filed by a licensed partner CPA.

The Tax Agent assembles your tax-year packet from live bookkeeping data, drafts the correct forms with reasoning-trace citations to the authorities, and hands off to a partner CPA who reviews, signs, and files. Quarterly estimates calendared and drafted for every customer — not a top-tier add-on.

Tools

Tax-year packet assembler
Form drafting (1120, 5472, 1120-S, 1040-NR, UK CT600)
CPA handoff — licensed partner reviews, signs, and files

Escalation

Every return is reviewed and signed by a partner CPA before filing. Complex positions (transfer pricing, PFIC, foreign tax credits) flagged for senior review.

Out of scope

State-level income tax returns (handled separately). Personal tax returns beyond 1040-NR. Audit representation.

Full Tax Agent page →

Agent 07

Compliance

Every deadline, calendared + filed

Cost budget (p95)

$0.50 per task (p95)

Model

Claude Haiku 4.6 — Sonnet for jurisdiction edge cases

Outcome

Zero missed filings across every jurisdiction your company touches.

The Compliance Agent runs a jurisdiction rules engine covering WY, DE, FL, CA, NY, TX, UK, IFZA, RAK, and Meydan. It auto-calendars every recurring obligation (annual reports, franchise tax, RA renewals, UK Confirmation Statement, UAE trade license), auto-files where APIs exist, and queues pre-filled drafts for one-click submission where they don't. US-formed entities no longer file a FinCEN BOI report — a final rule effective 14 August 2026 exempts them — so the agent does not calendar one.

Tools

Jurisdiction rules engine
State annual-report filing APIs
State annual-report APIs (where available)
Companies House confirmation statement filing
UAE free-zone renewal portals
Compliance dashboard with RAG status

Escalation

Filing rejections, jurisdiction changes, and non-standard situations (late fees, reinstatement, strike-off risk) route to ops.

Out of scope

Securities compliance (Reg D, Form D). Employment-law filings (EDD, 1099-NEC). Industry-specific licensing (healthcare, finance).

Full Compliance Agent page →

Agent 08

Co-pilot

Ask anything, with citations

Cost budget (p95)

$0.15 per message (p95)

Model

Claude Opus 4.7

Outcome

Answers to any question about your company grounded in your actual documents, transactions, and filings — with every claim citing its source.

The Co-pilot Agent runs chat over your full organization context — documents, transactions, filings, prior agent runs, prior conversations — grounded in a per-org knowledge search over your own records. Every answer includes inline citations linking to the source document, ledger entry, or filing. If it can't cite, it doesn't assert.

Tools

Per-org knowledge search
Document retrieval + reranking
Ledger query interface
Prior AgentRun search
Citation chip renderer

Escalation

Questions that would require legal or tax advice, or that touch policy decisions (e.g., 'should I move to a C-Corp?'), surface a suggestion to book time with a human advisor.

Out of scope

Opinions presented as facts. Hallucinated citations. Cross-organization data leakage (every query scoped by org).

Full Co-pilot Agent page →

How we prove it

Evals, budgets, audit trails.

Every agent ships with ≥20 eval cassettes — real recorded LLM traces replayed in CI on every change. New prompts have to beat the cassette baseline before they deploy.

Every agent has a cost budget asserted in tests. If the p95 cost of a run exceeds the published number, the build fails. That's how we keep the unit economics honest over time.

Every LLM call is logged as an AgentRun with prompt, tools, output, cost, latency, and outcome. Customers can export their own agent history. Ops can replay failures. Admin can set kill switches per agent, per jurisdiction.

See them actually running.

Start your company and the agents go to work.

Start your company Book a live demo