The semantic ledger. Built from scratch for discipline-as-data accounting — AI-augmented, jurisdiction-pluggable, audit-grade by construction.
Rules arrive as signed artifacts from tax authorities and standards bodies. AI proposes journal entries with reasoning. Deterministic processes do the math. Humans approve. Every entry is cryptographically chained, every state reconstructible at any past date.
Most "accounting software" today is a bookkeeping engine with reports bolted on. The accounting discipline — the rules, the reasoning, the regulatory chain — lives in the accountant's head, in external tools, in spreadsheets at year-end.
Sledger is built on the opposite commitment. The discipline is data the system loads: signed rule packs from tax authorities and standards bodies. The engine is generic. Add a jurisdiction by downloading its pack. Tax law changes? Authority publishes the new version, every conformant runtime picks it up by the effective date.
A row in a journal table. Amount, account, date. Why it happened, which contract it satisfies, whether the tax treatment was correct — implicit, ambient, in someone's memory.
A semantic event linked to its evidence, classified through reasoning, computed by signed authority rules, projected into the ledger. Audit becomes "check the reasoning" rather than "check the records."
Rules baked into vendor code. Updates ship on the vendor's release cadence. The customer's regulatory clock binds to the vendor's release clock.
Rules are versioned, signed UAPF artifacts published by the authority. New version goes live on its effective date. No software release on the regulatory critical path.
A complete accounting system needs more than a faster ledger. Sledger is six architectural commitments at parity, not a bookkeeping engine with features bolted on.
Tax rules, accounting standards, statutory methodologies arrive as signed UAPF artifacts. Versioned, effective-dated, executable. The engine never bakes them in.
Every event signed, content-hashed, linked to its evidence. Append-only, bitemporal. Tampering is mechanically detectable, not forensically argued.
AI never executes math that has a legal answer. It classifies, extracts, suggests — with confidence. Deterministic processes do the math. Humans sign. The chain records all three.
Adding a country is loading its rule pack. Latvia, Estonia, Lithuania, IFRS — all run on the same engine, side by side. Multi-framework reporting from one record.
Time-travel queries. Provenance traversal from any entry to AI proposal, rule pack version, evidence, approver. Hash verification independent of the system. Process re-execution against captured inputs.
UBL, Peppol BIS, EN 16931 for invoices. ISO 20022 for banking. SAF-T for audit data. XBRL for disclosure. OAuth and MCP for access. Everything interoperates because everything follows the standards.
An event arrives — a vendor invoice via Peppol, a bank line, a payroll trigger, a contract. Five steps later, a signed journal entry sits in the ledger with its full reasoning chain attached.
Input adapter normalises. Source document hashed and stored immutably in the evidence vault.
Frontier-class model classifies the event, extracts fields, identifies which rule packs apply. Confidence recorded.
Deterministic execution of the selected UAPF processes. Same inputs, same outputs, every run. Reasoning trace captured.
Approval queue surfaces the proposal with reasoning + evidence. Approver becomes auditor instead of data-entry clerk.
Append to the event store with full chain. Projections refresh. Submissions queued where obligation thresholds are met.
Fuzzy work — classification, extraction, anomaly detection. Confidence-scored. Never executes math with a legal answer.
Exact, law-bound, reproducible. Lives in signed UAPF processes published by tax authorities and standards bodies.
Reviewing reasoning chains, not entering numbers. Method, policy, exception, sign-off — the layer machines should not own.
"Add AI to QuickBooks" or "add AI to SAP" gives you AI-assisted bookkeeping. Different category. The architectural commitments below require the data model to be redesigned — once you do that, you are building from scratch.
Sledger does not invent its own protocols where mature standards exist. Everything that connects the system to authorities, banks, counterparties, agents already has a public spec. Sledger speaks them all.
EN 16931 compliant. Mandatory across the EU regulatory wave.
CAMT.053 statements, PAIN payments, native PSD2 API access.
Standard Audit File for Tax. Adopted in ~20 countries; auditor-ready out of the box.
Mature taxonomies for financial reporting. Listed-company-grade outputs.
Modern auth, with verifiable credentials for cross-organisational trust.
MCP-first. Agents are first-class operators, not API scrapers.
Universal Algorithmic Process Format. The signed, versioned rule artifact format Sledger reads natively.
Four-corner delivery network for cross-border B2B and B2G transactions.
Sledger is being designed openly. The memorandum, the system reference, the working notes — all public, all versioned, all welcome to be challenged.
The foundational document. What accounting is, who it serves, the three rule strata, the qualities of completeness. Technology-neutral by construction.
Read the memorandumThe architectural map. Twelve modules, five physical data stores, the rule lifecycle, the auditor interface, eight non-functional requirements. Companion to the memorandum.
Read the referenceWorking notes on UAPF, on rule registries, on Latvian accounting practice, on the gap between ERP-class systems and full discipline coverage. Iterative, in-progress.
Browse notesBuilding the next-generation accounting system. If you operate books, run an accounting practice, sit at a standards body, or want to put a rule pack into the registry — talk to us.