For developers

The financial brain
of the agentic economy

A deterministic financial engine your AI agents call over MCP, REST or API. The output is a certified three-statement model in which every figure is traceable. No hallucinated arithmetic.

Free sandbox · MCP-compatible · first call in minutes

The problem

LLMs are bad at numbers. Your users will notice.

  • LLMs are probabilistic text generators, not calculators; arithmetic and financial logic are structurally unreliable in them.
  • An ad-hoc calculation in a sandbox gives you a number, not a linked and verified model: the balance sheet doesn't tie, there's no audit trail and no industry methodology.
  • Financial output from an agent won't reach production without an audit trail (Gartner: >40% of agentic projects will be cancelled by 2027).

What you get

Not a calculator. A certified model — over an API.

A certified model, not a calculator

A three-statement model (P&L, cash flow, balance sheet) that ties, with an integrity certificate on every build.

Determinism and traceability

One input → one reproducible, auditable output. Every figure traces back to a formula and a source.

Multi-industry plus export

Real estate, SaaS, operating businesses; investment and operating models. Export with live formulas.

Integration

Three ways to connect

MCP — native tools

finmodel_build_from_brief (brief → certified model) and finmodel_chat — conversational refinement via session_id, synchronously, right inside your agent's tool loop. Plus read/export tools. MCP is already the standard: OpenAI, Google, Microsoft; 10,000+ servers; under the Linux Foundation.

REST — the same, asynchronously

POST /api/agent/build and POST /api/agent/chat → 202 {job_id}, poll GET /jobs/{id}. For headless agents and long-running builds.

fmk_ API keys

Scoped tokens (least privilege, RFC 8707 audience binding), workspace-level isolation.

MCP — synchronous, in the tool loop
# brief → certified model
finmodel_build_from_brief(brief="A coffee chain, 8 locations, seasonal demand")
{ model_id, session_id, certificate: "certified", export_url }
finmodel_chat(session_id, "add a downside scenario")
REST — asynchronous
POST /api/agent/build Authorization: Bearer fmk_…
→ 202 { job_id }
GET /api/jobs/{job_id}
{ status:"done", model_id, certificate, export_url }

All that leaves your system is a plain-language brief and a certified artifact. Your agent receives a finished, fully linked model and refines it conversationally — without assembling the financial logic itself and without risking a wrong number.

A number is not a model

“Why not just let the agent compute it in code?”

An agent in a sandbox can compute a percentage or a DCF. But building a linked, balanced, industry-specific three-statement model with an audit trail in ad-hoc code is expensive, slow and unreliable. We ship that as a service with an integrity guarantee: if it doesn't tie, the job says so honestly instead of sending broken numbers to your user.

Where it fits

  • AI CFOs and financial copilots
  • Vertical fintech agents (lending, underwriting)
  • Investment-analysis agents (deal screening, IRR/NPV)
  • Embedded financial planning inside vertical SaaS

Production-ready

Trust and standards

  • The MCP standard (OpenAI / Google / Microsoft, Linux Foundation)
  • Determinism plus an integrity certificate on every build
  • Scoped API keys from your dashboard, workspace isolation
  • Reproducibility and a complete audit trail

Price

Pay per call

A free sandbox to integrate against. After that, usage-based pricing (per call or credits) as is standard for APIs, with a hybrid base-plus-usage plan at volume.

Get an API key

Give your agents the financial brain