Infrastructure + open research · verifiable intelligent systems

Build autonomous systems that can prove they worked.

For teams that let AI agents write code, change production or act on real systems. Aftergraph binds agent work to scoped authority, durable execution and evidence that an independent verifier checks, so agent says done becomes verified outcome.

Verify this site live Talk to us Explore the platform View source
system lifecycle · aftergraph-siteVerified
Intent — A human or system declares purpose. Nothing runs without a declared why.
34canonical repositories
16public · 18 private
4public entry surfaces · honest maturity
7stages, intent to verified outcome
checking system…

Who it is for

Agent says done. Aftergraph checks.

Teams already let agents write code, change systems and run long jobs. The hard part is knowing the work actually happened, within the permission it was given. Three places Aftergraph is built for today.

Code review · prototype

Agent-written pull requests

Coding agents open PRs faster than people can review them. Sentinel evaluates the exact commit and returns a SHIP / DO NOT SHIP verdict with the evidence attached.

See Sentinel
Authority · governed

Agents acting on real systems

An agent that can deploy, pay or email needs scoped, expiring permission, not a shared key. AIE and Trust Gateway grant authority per mission and fail closed.

See Trust Gateway
Execution · durable

Long-running agent work

Multi-hour jobs crash, time out and restart. WORKS keeps leases, state and recovery, so the result can be traced back to every step that produced it.

See WORKS
Early teams

Building something similar?

Tell us what your agents do and where verification breaks today. Concrete failures shape the roadmap more than feature requests.

Start a discussion

Developers

Verify what is live — right now.

No signup, no key. The site proves which code it runs. Copy, paste, compare the SHA. Example output shown.

aftergraph.org/healthzprovenance · live
curl
curl -s https://aftergraph.org/healthz
# {"status":"ok","sha":"66b29af",...}
git ls-remote https://github.com/Aftergraph/aftergraph.org HEAD
# SHAs must match — live code equals reviewed code
JavaScript
const live = await fetch('https://aftergraph.org/healthz')
  .then(r => r.json());
console.log(live.status, live.sha); // ok 66b29af

What do you want to do?

Five ways into Aftergraph.

Choose by intent. Each entry path routes to an existing product, governed boundary, or evidence source.

Build

Compose governed intelligent work through the primary experience and developer surfaces.

Govern

Inspect authority, admission, policy, and canonical platform boundaries.

Execute

Follow durable work and runtime orchestration without collapsing their ownership boundaries.

Verify

Inspect evidence and independent verdicts bound to exact subjects.

Research

Read reproducible research and canonical architecture without promoting either into runtime authority.

Products

Four public entry surfaces. Honest maturity.

Studio, Wie, Sentinel and Atlas are the current public entry surfaces. The wider canonical portfolio is larger and includes private products, operator projections, capabilities, assurance and operations systems.

Wie by Aftergraph

prototype

Work Intelligence Engine. Source-neutral observations become structured, attributable WorkItems ready for review.

Signal → Observation → WorkItem → Review → Publish
View source

Sentinel by Aftergraph

prototype

Verified code review. PRs evaluated against exact HEAD with evidence-backed SHIP / DO NOT SHIP verdicts.

Exact HEAD → Checks → Evidence → Verdict
Sentinel details

System portfolio · Governance topology/2.0

The platform is larger than the front door.

Governance evidence cut 2026-09-30 records 34 canonical repositories. Visibility and lifecycle are shown separately from product maturity; topology membership does not imply production readiness. See all 32 systems.

34canonical
32permanent
30active
16public
18private

FIHIM

private repopersonal agent product

Persistent personal intelligent-agent environment and operator cockpit that composes canonical Aftergraph domain systems without taking ownership of their truth.

Recent signal · Economic Graph v1 operator projection · ad7e841
Inspect in Atlas

RenOS

private reposervice operations

Branded service-operations operator experience over Business Ops and tenant-specific domain truth, with attention and verification projections.

Recent signal · evidence-backed T24 staffing signal · a4a013a
Inspect in Atlas

War Room

public repoops intelligence projection

Organization-reality and operational-intelligence projection. It is a secondary diagnostic/backend surface — not the canonical operator/control owner.

Boundary · observe and correlate; never invent canonical state.
View sourceAtlas

CORE / ToolFabric

private repocapability routing

Capability-resolution and tool-routing control plane. Agents request capabilities; authority, secrets and execution remain owned by their canonical systems.

Recent signal · Labs Fabric v0.6 manifest-driven preflight · 86e5fd8
Inspect in Atlas

Skill ABI

public repoportable skills contract

Semantic compatibility contract for portable agent skills across models, runtimes and capability environments. Portable does not mean merely parseable.

Recent signal · ToolFabric binding constraints · aeccfe5
View source

Cron Fabric

public reposcheduled observation

Read-only scheduled organization sensing with evidence gating, dedupe and escalation. Observation explicitly grants no execution authority.

Recent signal · live org-pulse smoke: 27 discovered repos · 598fa98
View source

Catalog rule: canonical != public != production. Inspect the machine-readable catalog for all repositories, roles, visibility and lifecycle.

Open platform catalog

Agent–user boundary · illustrative

Stream the interaction. Govern the action.

Modern agent interfaces need ordered, inspectable events. Aftergraph keeps that interaction stream separate from authority, execution evidence and independent verification.

ACTIVITYStructured progress via snapshot/delta. progress ≠ completion evidence.
SUBAGENTProducer attribution and lineage across concurrent work. provenance ≠ ownership.
INTERRUPTExplicit pause/resume correlation for human or policy input. resume ≠ ambient permission.
Design directionEvent-stream UX inspired by open agent-interface patterns; not presented as an implemented AG-UI compatibility claim.Read machine contractRelated protocol: AG-UI

Why Aftergraph

Complete is not verified.

The system is designed around durable principles instead of confidence theatre.

Agent says done≠Verified outcome

Finished work still needs independent checking. Aftergraph separates self-report from proof at every layer.

1Claim — Agent reports completion
2Evidence — Inspectable artifacts emitted
3Independent verifier — Separate process checks criteria
4Verdict — Bound to exact subject, not self-report

Live evidence

Do not trust the claim. Inspect the system.

This surface cross-checks Aftergraph's same-origin health and deploy-provenance contracts at runtime. A verified state requires the exact SHA, route and deployment marker to agree.

Runtime probeaftergraph.org /healthz
Checking runtime
Healthunverified
Exact deploy SHAunverified
Deploymentunverified
Route contractunverified
Open public status Inspect raw health Inspect provenance Open exact commit Inspect Atlas View source

The whole company

All 32 Aftergraph systems, one map.

Every permanent system in the governance catalog, generated from /platform/catalog.json. Public systems link to source. Private systems are listed so you can see the shape of the company, and their code stays private. Being on the map does not mean production ready: lifecycle is shown on each card.

Products7

What people use: operator, personal-agent, work-intelligence, review and service-operations surfaces.

Relay

privatehuman operator plane

Governed operator plane for supervising autonomous work across agents, missions, execution nodes, Runtime projection, and Aftergraph infrastructure.

In Atlas

Studio

publicprimary experience

General-purpose human operating environment for Chat, Work, Space, control and evidence surfaces.

Source

FIHIM

privatepersonal agent product

Persistent personal intelligent-agent environment: identity/context continuity, agent coordination, user-facing conversational and delegated-work experiences, surfaced through HomeOS and other approved clients.

In Atlas

War Room

publicoperational intelligence backend projection

Operational intelligence projection, organization reality queries, missions/agents/compute/attention/evidence aggregation and secondary diagnostic surfaces without inventing canonical state.

Source

Wie

privatework intelligence experience

Wie browser experience and least-privilege BFF consuming canonical Work Intelligence state from wi-backend.

In Atlas

RenOS

privateservice operations product surface

RenOS branded service-operations operator experience, workspace composition, chat projection, attention/control presentation and domain/process/verification views.

In Atlas

Sentinel

publicverified code review

Exact-HEAD verified code-review verdicts with stale-base invalidation and cited evidence.

Source
Platform planes5

Authority, enforcement, runtime and durable execution underneath every product.

AIE

publicnormative authority

Portable institution, authority, delegation, lifecycle, budget and revocation semantics.

Source

Trust Gateway

publicruntime enforcement

Fail-closed runtime admission, policy enforcement, approvals, secrets and action audit.

Source

Runtime

privateagent runtime

Agent lifecycle, orchestration, dispatch, checkpoints, metering and observability.

In Atlas

WORKS

publicdurable execution

Durable work state, scheduling, workers, leases, recovery, execution evidence and quittance.

Source

Wie engine

publicwork inference

Source-neutral observation to canonical WorkItem inference, review, publication and promotion boundaries.

Source
Capabilities and research11

Continuity, skills, models, scheduling, knowledge and the research that keeps claims honest.

Context Continuity

privatecontinuity contract

Portable transfer of actionable context/state across model, agent, session and runtime boundaries.

In Atlas

Continuum

privatecontinuity containment verification

Continuity and containment fault-injection campaigns and verification harnesses.

In Atlas

Intelligence Systems Research

publicresearch assurance

SPEC-001, MISSION-Bench methodology, scientific claims, experiments, assurance and publication evidence.

Source

Skills Vault

privatecapability supply chain

Governed skill discovery, trust, lifecycle, provenance, compatibility and distribution.

In Atlas

LLM R&D

privatemodel rnd methodology

Reusable model research, experiment, evaluation, promotion and provenance methodology.

In Atlas

AFM

privatemodel program

AFM-specific model training, datasets, experiments, evaluations and artifact manifests.

In Atlas

Model Registry

privatemodel lifecycle registry

Immutable promoted model identities, versions, aliases, lifecycle, provenance and artifact locations.

In Atlas

Cron Fabric

publicscheduled observation fabric

Read-only scheduled organization sensing, evidence gating, dedupe and Telegram escalation. It grants no execution authority.

Source

Docs

publicknowledge plane

Provenance-pinned rendering, discovery, developer documentation and agent-readable context surfaces.

Source

Skill ABI

publicskill compatibility contract

Semantic compatibility, bounded effects, degradation and conformance for portable AI agent skills.

Source

CORE / ToolFabric

privatetool routing control plane

ToolFabric registry federation, semantic capability resolution, health-aware routing plans and non-authoritative tool selection.

In Atlas
Foundation, governance and domains9

Contracts, brand, community, the front door, business domains and systems in transition.

Governance

publiccanonical contracts

Platform topology, cross-repo boundaries, contract registration and generated org-state mechanics.

Source

AVC

privatelegacy migration sourcelegacy transition

Legacy product, Hermes integration and migration-source behavior pending governed extraction.

In Atlas

Veranza

privateassurance incubationinternal hold

Internal assurance-product incubation concept under naming/clearance hold. Topology membership does not imply public or production maturity.

In Atlas

aftergraph.org

publicpublic front door

Public website, marketing information architecture and system launcher.

Source

Brand OS

publicbrand design system

Visual identity, semantic design tokens, master assets and public communication design rules.

Source

Community (.github)

publicorganization community

Organization profile, contribution defaults, security/support routing and shared community infrastructure.

Source

Skillport

privateskill portability benchmark

Open benchmark and evaluation suite for semantic portability of skills across models, runtimes and capability environments.

In Atlas

Business Ops

privatecanonical service business domain

Canonical tenant-neutral service-business domain state, invariants, migration mappings, charge-source facts and shadow-conformance semantics.

In Atlas

Rendetalje

privaterendetalje reference tenant

Rendetalje-specific cleaning semantics, pricing policies, estimator behavior, playbooks, communication policy, tenant configuration, migration mappings and tenant-specific Business Ops composition.

In Atlas

Platform · Illustrative system walkthrough

Seven permanent planes. One verifiable lifecycle.

V4 keeps semantic ownership explicit. Evidence and Verified Outcome are lifecycle results — not extra planes.

ExperienceStudio→ IntelligenceWie→ AuthorityAIE→ TrustTrust Gateway→ RuntimeRuntime→ ExecutionWORKS→ VerificationSentinel

Lifecycle: Intent / Experience → Intelligence → Authority → Trust → Runtime → Execution → Evidence → Verification → Verified Outcome.

Architecture source
Inspect in Atlas

Evidence

Research is not runtime authority.

A claim can be useful without being promoted. Every surface tells you what kind of evidence you are looking at.

01 · Result

Research result

SPEC-001 and MISSION-Bench make research claims reproducible and inspectable.

02 · Contract

Specification

Governance defines boundaries and contracts without silently becoming implementation.

03 · Reality

Implementation

WORKS, Trust Gateway and other repos expose what is actually built.

04 · Verdict

Verified production

Only independent evidence can justify a stronger operational claim.

Research evidence can justify a promotion decision; it is never runtime authority by itself.

Inspect evidence

Principles

Durable by design.

01Purpose-bound authority

Permissions are scoped to the mission, never ambient.

02Durable execution

Long work survives interruption with leases and recovery.

03Deterministic gates

Claims point to inspectable artifacts, not opinions.

04Cost per outcome

Efficiency measured against what was proven.

05Honest labelling

Prototype, research and production stay visibly distinct.

History

Shipped in the open.

Merged history plus the current branch direction. Every production entry is inspectable on GitHub.

2026-10-01Platform truth refresh — Governance topology/2.0, 34 canonical repos, live deploy provenance and agent interaction boundaryfrontier
2026-09-14V3 product-first front door — Studio, Wie, Sentinel and intent router (#72)production
2026-09-10Atlas interaction layer — evidence-aware observatory, agent contract, gated deploys (#58–#70)production
2026-09-09Brand OS v1.1.0 — canonical visual system adopted across the site (#65)production
2026-09-08Topology reconciled — public surface aligned with Runtime and Wie truth (#53)production

Inspect it before you trust it.

Explore the system topology, read the contracts, inspect evidence, then launch governed work.

Open Atlas Read the docs Browse GitHub Talk to us Open launcher
34 canonical repositories · 16 public · 18 private30 active · 32 permanent · 2 temporaryResearch evidence is never runtime authority