Every agent remembers differently
Private per-agent context produces silos. Teams need shared evidence that survives tool and model changes.
Typed evidence infrastructure
Institutional memory for agent-native enterprises
Ontic Fabric gives people, agents, and auditors one governed substrate for knowledge: typed ontologies, provenance-bearing evidence, durable work streams, and document and audio transduction.
Agents change. Your intelligence remains.
Models, tools, and sessions are replaceable. Decisions, obligations, code, documents, and evidence remain addressable in your substrate.
The enterprise problem
Agent productivity without memory becomes institutional loss
When agents write code, review contracts, resolve incidents, and summarize decisions, their work creates organizational knowledge. If that knowledge lives only in chats, tickets, or private sessions, it disappears when the session ends.
Plausible text is not evidence
An answer needs a source, a time, and a type. Ontic Fabric preserves provenance and graph relationships instead of loose summaries.
Work streams outlive sessions
Intent, dependencies, approvals, validation, and outcomes become typed records that the next agent—or auditor—can traverse.
The substrate
One typed evidence layer across every domain
Ontic Fabric is not another workspace application. It is the memory and evidence layer underneath the tools your organization already runs.
- Ontologies as configuration. Define the objects, relationships, markers, and ingestion routes for software, legal, commerce, infrastructure, and operational work.
- Provenance by default. Source files, derived evidence, sessions, and graph edges retain lineage.
- Filesystem-native access. Linux agents can work with memory through ordinary filesystem semantics; other clients reach the same tree through supported gateways.
- Transduction. Bring PDFs and declared audio evidence into the substrate; published matrices will state fidelity, latency, resource use, and cost.
- Portable sovereignty. Your content and basic graph organization are exportable by design.
# one substrate, several typed domains legal.obligation ob-104 └─ satisfied_by ─▶ service svc-auth service svc-auth ├─ implemented_by ─▶ repository repo-identity └─ changed_by ─▶ work_item work-8841 work_item work-8841 ├─ validated_by ─▶ test_run run-2026-11-13 └─ approved_by ─▶ reviewer security-team
Governed work streams
Turn agent activity into defensible operational records
Ontic Fabric represents work as a typed graph: intent, dependencies, assignments, evidence, reviews, and outcomes. Humans keep approval authority while agents execute against explicit scope.
Capture intent
Define the outcome, constraints, owners, and acceptance evidence before dispatching expensive work.
Connect execution
Agents, repositories, documents, and validation results remain linked instead of scattered across logs.
Prove outcomes
Reviews reference tests, deployments, approvals, and source evidence—not unsupported claims.
Retain knowledge
Reusable skills and session fossils carry forward what worked, with provenance back to the originating work.
Empirical evolution
Improve agent lanes with measured evidence
A lane combines a scope, ontology, skill set, model policy, context budget, and evaluation set. Benchmarked lane versions make agent operations improvable rather than anecdotal.
Capture what actually happened
Sessions, work items, reviews, and outcomes become substrate records.
Benchmark focused lanes
Measure task success, evidence quality, review load, inference usage, and escaped defects.
Version what works
Promote skills, fossils, and lane definitions with their evaluation provenance.
Deployment
Sovereign locally. Governed at enterprise scale.
The same substrate runs on an individual workstation, an isolated managed workspace, or an enterprise-controlled deployment.
Local and sovereign
Run the substrate on hardware you control, with exportable content and no mandatory cloud dependency.
Operated for a team
An isolated workspace with deployment, upgrades, backups, drift monitoring, and support.
Your control boundary
Bring your own cloud or use an isolated managed tenant, with governance and audit controls matched to the engagement.
Assurance
Measured, not promised
We will publish the evidence buyers need to evaluate the substrate rather than asking them to accept adjectives.
| Commitment | What we will publish | Status |
|---|---|---|
| Retrieval quality | Reproducible recall and typed-evidence benchmark results, including baseline comparisons. | In preparation |
| Transduction | Supported format matrix with fidelity, provenance behavior, latency, resources, and cost. | In preparation |
| Security assurance | SOC 2 / ISO 27001 readiness scope and report status when issued. | Roadmap |
| Operations | Backup and restore drills, upgrade policy, incident history, and service limits. | Pilot documentation |
Closed pilots
Put your organization's evidence on one substrate
Pilots start with a bounded corpus, explicit acceptance evidence, and measured retrieval and transduction results. No public claims are made from your data.