OpenFlamma Docs
v0.1 · Preview Open site ↗
Developer preview

The open financial ledger for AI agents in global trade.

A shared financial layer for agents working across trade orders, documents, bank feeds and multi-currency payments—with human approval built in.

OpenFlamma is built around the operating reality of iBnk’s trade customers. AI agents are ready to match transactions, draft journal entries, check invoices and follow trade documents—but today’s finance systems give them no shared ledger, no clear authority and no durable record of what they did.

OpenFlamma gives them all three:

01

An open ledger

A double-entry ledger for fiat and digital assets. One source of truth agents can work from and your team can approve.

02

Permissions and policies

Simple rules define what each agent can read, what it can propose, and what needs human approval.

03

A complete audit trail

Every agent action is recorded: what it did, what data it used, who approved it, and when.

Primary users

Built around trade operations

↗

Importers & exporters

Following the documents, currencies and payments behind every cross-border order.

◎

Trade finance teams

Giving agents reconciliation and bookkeeping work while keeping approval authority.

{ }

Agent developers

Building agents that need safe access to transaction-level trade and financial data.

Deployment model

Open source and Cloud

The ledger, permissions and audit trail are open source. Run them yourself, inspect every line, and keep your data. OpenFlamma Cloud is the hosted version, with agentic reconciliation, a trade finance assistant, a console and support built in.

CapabilityOpen sourceCloud
Ledger, policies and auditIncludedIncluded
Where it runsYour infrastructureManaged by iBnk
Agent products and consoleBuild your ownIncluded

OpenFlamma is built by iBnk, which serves 1,000+ importers, exporters and trading companies.

Getting Started

Quickstart

Coming soon

The five-minute Docker guide will cover starting OpenFlamma locally, creating a ledger, recording a first balanced transaction and reading the resulting balances.

terminal
# Docker image and public repository will be published
# with the first OpenFlamma source release.

We do not publish placeholder commands that could create an incompatible setup.

Mental model

Core Concepts

These concepts describe how people, agents and financial evidence fit together. Names and relationships are stable; implementation details will be added as the source becomes public.

01

Ledger

The single record of every financial event. OpenFlamma uses double-entry bookkeeping: every transaction moves value from one account to another, so the books always balance. Entries are never edited or deleted, only corrected by new entries, so the history remains complete.

02

Accounts

A place where value is recorded, such as a bank account, a customer balance, a supplier balance, revenue or fees. Each account can hold several currencies and digital assets.

03

Transactions

A balanced set of entries across two or more accounts. A transaction can carry references to its source, such as an invoice, a bank statement line or a shipping document.

04

Agents

Any AI agent or automated program that works with the ledger. Each agent has its own identity, so every action can be traced back to it.

05

Policies

Rules that decide what an agent may read, propose or commit—and when a person must approve the action.

06

Human approval

When a policy requires a person, the agent’s work waits in a review queue. A person approves, edits or rejects it. Agents do the work; people stay in control.

07

Audit trail

A permanent log of every action by people and agents: who did it, what changed, what data was used, which policy applied and who approved it.

08

Sources

Where data comes from: bank feeds, payment records, invoices, contracts and trade documents. Every ledger entry can link back to its source.

Design constraints

Our Principles

  1. 01

    Open by default.

    Finance needs proof, not promises. The code that touches your books should be open to inspect.

  2. 02

    Agents do the work, people decide.

    Agents take on repetitive work. Important decisions stay with people.

  3. 03

    Every action is traceable.

    Nothing happens without a record of who did it and why.

  4. 04

    Your data stays yours.

    Self-host it, export it, take it anywhere. No lock-in.

  5. 05

    Built for global trade.

    Multiple currencies, digital assets and cross-border documents belong in the core, not an add-on.

Agents

Permissions, policies and approval

Every agent has an identity and a policy boundary. A policy can make an agent read-only, allow it to draft a specific action, restrict it by account or amount, or require a person before anything is committed.

A policy answers four questions

  • What can this agent read?
  • What can it propose?
  • What can it commit directly?
  • Which actions need a named human role?
conceptual policy · syntax not final
agent: reconciliation-assistant
read:
  - bank_feeds
  - invoices
propose:
  - journal_entries
approval:
  above: USD 10000
  role: finance_manager

This example explains the intended policy model. The final policy schema will be published with the API specification.

Agents

Audit Trail

The audit trail connects an action to its actor, evidence, policy and decision. Finance teams can move from a ledger entry back to the source documents and approval that produced it.

Recorded event
Actor
reconciliation-assistant
Action
Proposed journal entry
Evidence
INV-2034 · bank-line-8912
Policy
journals-over-usd-10k
Decision
Approved by finance_manager

Illustrative event only. Export formats and event field names will be documented with the first API release.

Reference

API Reference

Coming soon

The public contract will ship with the source release. No endpoint paths, package names or tool schemas are published yet.

REST

REST API

Ledger, accounts, transactions, policies, approvals and audit events—with request and response examples.

Specification pending
SDK

TypeScript & Python

Typed clients, authentication, errors and runnable examples for both supported languages.

Packages pending
MCP

MCP Server

A discoverable tool list for agents to read ledger state, propose actions and request approval.

Tool schemas pending
LedgerAccounts, transactions, multi-asset accounting and trade chart-of-accounts templates.Coming soon
IntegrationsBank feeds, invoices, documents and accounting exports.Coming soon
Self-HostingDocker, Kubernetes, configuration, security and backups.Coming soon
CommunityContributing, roadmap and licence notices will publish with the source repository.Coming soon