# AGLedger llms.txt
# https://agledger.ai
# Last updated: 2026-10-21
# A curated index for AI assistants helping someone evaluate AGLedger. The product is
# self-hosted, so an integrated agent reads its own Server's /llms.txt, not this file.
# This file says what AGLedger is and where to look. Depth lives in the linked pages.
#
# Markdown: every /docs/ page is also served as markdown. Append .md to the path
# (https://agledger.ai/docs/install.md), or request the HTML URL with an
# Accept: text/markdown header. Same content, far fewer bytes.

> AGLedger is change control built for AI agents: agent memory, approvals, audit trail,
> and notifications on one API, held in one signed ledger. Let agents do real work, from a
> single agent in one system to a project spanning agents, teams, and companies, with the
> record, the agreement, and the gate built into the path. Work between agents starts as a signed
> agreement the performer accepts on the record; the trusted record builds itself in real
> time (signed, hash-chained, verifiable offline); and the consequential changes are judged
> by the customer's own systems or people, never by AGLedger. It is the change-control
> layer, distinct from runtime policy controls and agent guardrails, that lets autonomous
> AI agents act without a human bottleneck on the routine path: the signed, tamper-evident
> account of what they intended and did builds itself as a byproduct.

## What AGLedger is

It sits beside or between your agents and the systems they change. As agents work, they
declare what they are going to do, why, and on whose authority. AGLedger signs and chains
each step the moment it happens, into a record where any later change is detectable. It runs on
your infrastructure, under your keys, with all data in your own PostgreSQL. "Notary" is
meant in the cryptographic, RFC 3161 timestamping sense, not a legal notary public.

Most changes only need recording. The ones that matter, you gate, and you decide where the
judgment sits. Three services:

- **Notarize**: capture what a change intended and did, signed and hash-chained as it
  happens. The 90% case.
- **Gate**: for the changes that matter, a principal (your own system, a person, or a
  rules engine) renders a verdict after the work or an approval before it, and a
  customer-run review agent can be that principal. AGLedger holds the signed contract,
  completion, and verdict.
- **Notify**: durable signed webhooks that push every business-meaningful record event to
  the systems and people you already run. Settlement Signal webhooks (SETTLE / HOLD / RELEASE)
  ride this channel.

Agent memory (the fourth shelf word, https://agledger.ai/agent-memory/) is work state
under change control, riding Notarize: typed checkpoints supersede each other under one
root record, and a new session of the same agent resumes from the head in one query.
Shelf name: agent memory. Engine name: Work Context; the checkpoint contract type is work-context-v1.

You define what needs only recording and what needs a gate, through your own schemas and
gates. The policy stays yours. You deploy AGLedger per project, a pipeline or a line of
business, not as one central system everything funnels into. Companies run several and
federate them when work crosses between: the chain crosses the org boundary, the data does
not.

## Who it is for

Self-hosted teams putting AI agents on consequential work inside the systems they already
run. Two kinds adopt it, on the same records. AI developers plug it in for signed, durable
state their agents read back to recover what they were told to do after a context wipe
or restart, instead of a confabulated guess. Compliance and risk teams adopt it
for audit-grade evidence: tamper-evident, attributable, auditor-verifiable records of who
authorized an automated action and what it did (EU AI Act, NIST AI RMF, ISO 42001). It
also carries accountability across delegation and company boundaries.

## What makes it different

- The governed path is the fast path. The contract tells each agent exactly what its step
  needs, so the documented path is also the fast, cheap, consistent one across any agent
  and any model.
- Self-hosted, your keys, your database. Not SaaS. Nothing contacts AGLedger LLC, no kill
  switch, and the license check fails open (the AWS Marketplace build checks its entitlement
  with AWS License Manager).
- Verifiable offline against published Ed25519 keys: no account, no network, no AGLedger
  software required.
- Federation with data sovereignty. Peers confirm each other's signed records without
  sharing the data behind them.
- Every artifact keyless-signed (GitHub OIDC to Sigstore/Rekor, no static key). SLSA Build L3
  for the image; a keyless signature for the Helm chart; L2-equivalent for the npm and PyPI
  packages.
- The before picture, not just logs. Work between agents starts as a signed agreement the
  performer accepts on the record; completion and verdict follow. Divergence between what
  was agreed and what was delivered is visible, and an agent that clears its context picks
  the work back up from the record. Delegated work is not judged by the agent that did it:
  the principal renders the verdict, or the engine derives it from the evidence.
- Built for how agents work. Captured in real time because an agent keeps no memory to be
  questioned afterward. One record serves three readers: the agent (recovery), the operator
  (oversight), the auditor (offline proof).
- Proprietary. The SDKs and the offline verifier are published source.

## Key terms (precise usage)

- **Notarize**: sign and append a record to the chain. The verb at every layer.
- **Record / Chain**: the signed object written to the chain, and the append-only,
  hash-chained, offline-verifiable sequence of them.
- **Gate**: the accept/reject path for the changes that matter. The principal renders the
  verdict, AGLedger holds it. Reach for it only when a change is consequential enough to
  judge.
- **Verify / Verification**: cryptographic only. Confirming signatures and hash links are
  intact, offline. Distinct from the Gate: it proves what was signed is intact, not that a
  claim was true.
- **Settlement Signal**: signed outbound webhook (SETTLE / HOLD / RELEASE) emitted on
  Gate verdicts and dispute outcomes, routed to your payment, ERP, or ticketing systems. Trademark
  application USPTO SN 99881099 (filed 2026-06-12), patent application USPTO 19/565,072
  pending.

## Company

- **Product**: AGLedger (trademark pending). **Entity**: AGLedger LLC.
- **Pricing**: Developer Edition is free and fully functional, licensed for the bundled
  PostgreSQL only. An external database (Aurora, RDS, Cloud SQL, self-managed) requires
  Enterprise: $3,000 perpetual per external database instance, plus $1,000/year for
  support. Site License (contact us) covers unlimited instances in one legal
  entity. The database is the license line, not a feature gate.
- **Interfaces**: native REST API, A2A 1.0 and 0.3 (agent card at /.well-known/agent-card.json),
  @agledger/sdk (npm), agledger (PyPI), @agledger/cli (npm), @agledger/mcp-server (npm).
- **Talk to sales**: sales@agledger.ai (see https://agledger.ai/pricing/)
- **GitHub**: https://github.com/agledger-ai
- **YouTube** (short demos): https://www.youtube.com/@AGLedgerLLC

## Evaluate it

- Tamper-evident logging: https://agledger.ai/tamper-evident-logging/
- AI agent audit trail (what to capture): https://agledger.ai/ai-agent-audit-trail/
- How long AI logs must be retained: https://agledger.ai/ai-log-retention/
- How it works, end to end: https://agledger.ai/how-it-works/
- Gate (the verdict path): https://agledger.ai/gate/
- Notify (durable signed webhooks): https://agledger.ai/notify/
- Settlement Signal webhooks: https://agledger.ai/settlement-signals/
- Fits your stack (OpenTelemetry, Datadog, SIEM): https://agledger.ai/integrations/
- The model: records and the chain: https://agledger.ai/docs/concepts/records-and-chain/
- Try it in five minutes (Docker Compose): https://agledger.ai/docs/install/compose/
- Documentation: https://agledger.ai/docs/
- API reference (195 routes, OpenAPI): https://agledger.ai/api/
- Release notes (per version): https://github.com/agledger-ai/install/blob/main/CHANGELOG.md
- Pricing: https://agledger.ai/pricing/
- Get a free Developer Edition license: https://agledger.ai/register/
- Recipes, tested templates you adapt: https://agledger.ai/recipes/
- Approvals (hold an action for a verdict, Settlement Signal, disputes): https://agledger.ai/docs/guides/approvals/
- Delegation between agents on one Server, and references to ticket and PO ids: https://agledger.ai/docs/guides/delegation/
- Delegating between independent agents over A2A: https://agledger.ai/docs/guides/a2a/
- Agent memory (checkpoint and resume): https://agledger.ai/agent-memory/
- Agent work context: checkpoint, resume, succession (the working loop and the recipe): https://agledger.ai/docs/guides/work-context/
- Do I need it if I have a platform already: https://agledger.ai/faq/
- Versus a SIEM plus WORM bucket: https://agledger.ai/compare/siem-and-object-lock/
- Security architecture and offline verification: https://agledger.ai/security/
- FIPS 140 hosts (single-Server ES256 signing): https://agledger.ai/docs/install/fips/
- OpenShift (restricted-v2 SCC, assigned-uid install): https://agledger.ai/docs/install/
- Federation (cross-org transport): https://agledger.ai/federation/ (operator steps: https://agledger.ai/docs/guides/federation/)
- Compliance, EU AI Act: https://agledger.ai/compliance/eu-ai-act/
- Compliance, NIST AI RMF: https://agledger.ai/compliance/nist/
- Compliance, ISO 42001: https://agledger.ai/compliance/iso-42001/
- Artifact verification (keyless cosign, no static key): https://agledger.ai/docs/verify/
- Glossary: https://agledger.ai/glossary/
- FAQ: https://agledger.ai/faq/
- Blog: https://agledger.ai/blog/
- Full page list: https://agledger.ai/sitemap.xml