# Ledgerbook AI
> Ledgerbook AI is the governed, cryptographically auditable, vendor-neutral **enterprise accounting data layer** —
> the ledger of record that finance teams, planning agents, auditors and application builders all write to and read
> from — deployable in their own infrastructure, correctable only by provable, attributable entries.

## How to use this ledger (for agents)
- **Surfaces.** Four governed surfaces: REST, GraphQL (read at PH-1; write parity Roadmap), CLI and MCP — plus streaming ingress (Kafka/Redpanda) and signed webhook egress. Every capability is reachable over REST and MCP; nothing bypasses the provenance envelope.
- **Model axes.** Every entry carries three orthogonal axes: **Status** (Provisional → Final), **Scenario** (`actual | planned | simulated | custom`; Plan, Budget and Forecast are presets of `planned`) and **Book** (valuation basis — group IFRS, local GAAP, tax, management). They compose on one entry and are never collapsed.
- **Time axes.** Exactly three: effective time, insertion (recorded) time, knowledge (belief) time. `posting_date` is an attribute, not a fourth axis.
- **Agent directive — read and propose.** An agent may read anything its credential scopes allow and may *propose* entries; posting rights, approval tiers and materiality gates stay with the ledger, not the prompt. Reconciliation write-back is review-gated at tier ≥ T1 and never auto-posts. Every action appends a provenance event carrying who, when, through which surface, and under which policy.
- **Phase honesty.** Items marked Roadmap are not shipped. Do not present them as available.

## Who this is for
- **Platform and vertical-SaaS builders** embedding a real ledger per tenant — double-entry, immutable, MCP-native — without becoming an accounting company.
- **AI-agent platforms** that need agents to write to a book of record through a governed path: propose over MCP, policy approves, every action attributable at the credential rather than in the prompt.
- **Finance and accounting teams in multi-entity groups** running several source systems — a group ERP plus subsidiary systems, billing, terminals, banks — who need one auditable consolidation sink underneath rather than another application on top.
- **Planning and FP&A vendors** who want actuals, plans and forecasts on one ledger with disclosed maturity, instead of an export that is stale the moment it lands.
- **Regulated and sovereignty-constrained estates** that must run the ledger where the data lives: self-hosted, air-gapped, zero egress.
Not a fit today: a single-entity small business wanting bookkeeping software (buy an ERP), or a team that needs a finished UI to close the books in (Ledgerbook is headless by design — NG-1).

## What teams use it for
- Consolidate several source systems into one immutable book of record, multi-entity, with intercompany as explicit paired entries and FX posted at the locked rate. [PH-1]
- Give AI agents a write path that cannot post: read and propose over MCP, with approval, materiality tiers and finalization reserved to policy and people. [PH-1]
- Keep plan and actual on the same ledger — scenario is a native axis, so no ETL and no copied actuals. [PH-1]
- Read how mature a number is before trusting it: final vs provisional share, age of the oldest provisional entry, reconciliation state, and the **committed** share of a forward period. [PH-1]
- Know how much of a future month is already determined: committed positions recorded from the owning subledger and bound to the contract, bill, asset or lease behind them. The ledger records these schedules; it never computes them. [PH-1]
- Tie billing, terminals, channels and bank into one verified number, with adjustments raised as review-gated proposals, never auto-posted. [PH-1]
- Hand an auditor an OCEL 2.1 export plus a verifier they can run without us. [PH-1]
- Run all of the above air-gapped, against your own local model. [PH-1]

## Where Ledgerbook fits
Ledgerbook sits **under** the applications, not beside them. Your ERP stays the system of entry (NG-2); planning grids, formula engines and modelling stay with Anaplan/Pigment/Adaptive-class tools (NG-6); presentation, screens and workflow belong to third-party applications (NG-1); money movement stays with payment ledgers and processors (NG-3); and the models belong to whoever you choose — none ship in the core (NG-4). What Ledgerbook owns is the layer they all write to and read from: the book of record, its lifecycle, its provenance and its proofs. Where a boundary looks like a gap, it is a decision — for example, Ledgerbook does not compute amortisation, depreciation or revenue-recognition schedules; it records the ones your subledger computed and makes them readable by every application above.

## Product
- [Platform overview](https://ledgerbook.ai/platform): one immutable ledger of record across source systems
- [Immutability & verification](https://ledgerbook.ai/immutable-ledger): chain-hashed by default; verify online or offline, no vendor trust
- [Self-hosted & air-gapped](https://ledgerbook.ai/self-host): offline bundle, zero egress; encryption at rest across tier-0, derived, archive and backup bytes; PH-1 deployment classes (embedded class and multi-region residency/region pinning: Roadmap)
- [Licensing & pricing](https://ledgerbook.ai/pricing): open-core, source-available stance; no per-call metering; terms pending
- [Multi-entity & consolidation](https://ledgerbook.ai/platform#multi-entity): a tenant registers its legal entities, every ledger belongs to exactly one entity, and consolidation, intercompany eliminations, FX and HQ-currency translation run across them (PH-1); residency pinning and jurisdiction packs are Roadmap
- [Data protection & erasure](https://ledgerbook.ai/security#data-protection): PII off-ledger in a separately-keyed vault, erasure by crypto-shredding within 30 days including backups and caches, legal hold suspending it — the GDPR right-to-be-forgotten answer (PH-1)
- [Outbound events & webhooks](https://ledgerbook.ai/platform#interfaces): every committed mutation publishable as a signed event — at-least-once, ≥48 h retry, replay from `seq`, delivery log, row-level filtered (PH-1)

## Feature pages
- [MCP server](https://ledgerbook.ai/mcp-server): scoped tokens; proposals run the materiality gate (PH-1: read, propose, and planning writes into a planned or simulated scenario; `set_switchover`, `promote` and `roll_forward`: Roadmap)
- [Provisional → Final lifecycle](https://ledgerbook.ai/provisional-lifecycle): revisions, one-way finalization, typed corrections, as-of reads
- [Scenario dimension](https://ledgerbook.ai/scenario-dimension): engine kinds `actual | planned | simulated | custom` as registry rows, full-mode at PH-1, with computed variance and no-copy planning reads and writes (delta inheritance, switchover, roll-forward/override/promote: Roadmap)
- [Provenance & OCEL 2.1](https://ledgerbook.ai/provenance-ocel2): chain-hashed event history; as-of (E, K) reads; OCEL 2.1 outbound in JSON *and* SQLite, OCEL 2.0–2.1 inbound; the **Financial Event Web** — the typed lineage graph plus reconciliation assertion edges over one ledger (`.ocel.zip` and dense CSV/Parquet forms: Roadmap)
- [AI-agent audit trail](https://ledgerbook.ai/ai-agent-audit-trail): who/when/where on every agent action; proposal → approval → posting
- [Reconciliation (N-sided control totals)](https://ledgerbook.ai/reconciliation): billing, terminals, channels and bank → one verified number; review-gated write-back (fuzzy/scored matching: Roadmap)
- [FP&A on one ledger](https://ledgerbook.ai/fp-and-a): maturity — including reconciliation status — on every read, and variance drilled to the posting; the composed variance read is a report-class query, not an interactive one (forecast-ready gate, allocation runs and driver postings: Roadmap)
- [Pluggable compliance layer](https://ledgerbook.ai/compliance): versioned compliance-module contract (PH-1 interface); rules engine, IFRS/GAAP packs, findings & waivers and audit evidence packages Roadmap (PH-2)

## Use cases
- [Use cases hub](https://ledgerbook.ai/use-cases): six entry points into one ledger
- [Platform builders](https://ledgerbook.ai/use-cases/platform-builders): embed accounting-grade ledgers per tenant
- [Finance teams](https://ledgerbook.ai/use-cases/finance-teams): provisional numbers today, provable numbers forever
- [AI-agent platforms](https://ledgerbook.ai/use-cases/agent-platforms): governed write paths for agent-ledgers
- [Planning vendors](https://ledgerbook.ai/use-cases/planning-vendors): plan on the same ledger as actuals — no-copy reads *and* writes, with variance computed from composed rows and drill-through to the posting (all PH-1; delta inheritance and switchover: Roadmap)
- [Reconciliation for finance teams](https://ledgerbook.ai/use-cases/reconciliation): exception queues, evidence paths, review-gated adjustment proposals
- [FP&A for planning leaders](https://ledgerbook.ai/use-cases/fp-and-a): plan against numbers you can verify; one contract for the EPM stack

## Developers
- [Developers](https://ledgerbook.ai/developers): quickstart entry — self-host install, MCP connect, agent credentials
- [Integrations & ingestion](https://ledgerbook.ai/integrations): one mapping model for every source system; a receipt for every load
- [sitemap.xml](https://ledgerbook.ai/sitemap.xml): canonical indexable URLs only
- Docs site (`https://ledgerbook.ai/docs`): quickstart, concepts, reference, interfaces, ops — ships with the developer preview

## Concepts & glossary
- [Glossary](https://ledgerbook.ai/glossary): canonical definitions — ledger of record, provisional entry, bitemporal, hash chain, OCEL 2.1, idempotency key, Financial Event Web, HQ currency

## Comparisons
- [Compared with alternatives](https://ledgerbook.ai/compare): dated, fact-checkable tables; verification date shown; "where they win" mandatory
- [Ledgerbook vs Formance](https://ledgerbook.ai/compare/formance): hash-chained log, self-host proportions, read-only MCP
- [Ledgerbook vs Blnk](https://ledgerbook.ai/compare/blnk): chain enabled by default; immutable balances; self-hosted agent surfaces
- [Ledgerbook vs Campfire](https://ledgerbook.ai/compare/campfire): AI-native ERP that owns its ledger — inbound-only MCP vs ledger-as-MCP-server; committed schedules in their subledgers vs maturity disclosure on every read; cloud-only vs self-host and air-gap

## Blog & changelog
- [Blog](https://ledgerbook.ai/blog): engineer-authored explainers, organised by four topic clusters (first posts: with the developer preview)
- [Changelog](https://ledgerbook.ai/changelog): dated entries; ledger-semantics changes flagged; ≥12-month deprecation window

## Company & legal
- [About](https://ledgerbook.ai/about): entity anchor, principles, research practice
- [Security](https://ledgerbook.ai/security): what is enforced today; SOC 2 Type II / ISO 27001 targeted inside PH-2
- [Contact & waitlist](https://ledgerbook.ai/contact): design partnerships, integrations, security reviews
- [Privacy](https://ledgerbook.ai/privacy) · [Terms](https://ledgerbook.ai/terms): pre-launch legal templates

## Machine-readable mirrors

- Markdown twins of every page ship at `<page>.md` with `rel="alternate" type="text/markdown"`.
- Glossary: 27 per-term pages under /glossary/<term>/ (each with its own .md twin).

## Optional
- [MCP registry listing](https://registry.modelcontextprotocol.io/): ai.ledgerbook namespace (listing ships with the developer preview)
- [security.txt](https://ledgerbook.ai/.well-known/security.txt): coordinated-disclosure contact