Source & scope discovery
We inventory the data sources that already exist, learn their semantics, and define how access should be gated, all before a line of integration is written.
Plabs / Products / Intelligence Fabric
Flagship platform Bespoke engagement Enterprise-scale
The connective tissue between your data and the AI you actually want.
Intelligence Fabric is a bespoke intervention, not a SaaS. We discover your data sources, gate access, and expose them as a governed tool suite over OpenAPI and MCP, with prebuilt orchestrators, hot-path detection, automatic crosswalks, and an audit trail your security org will sign off on.
01 / Architecture
The shape of it
A governed surface across every system you already own.
FIG. 01 / FABRIC TOPOLOGY v.2026.04
CONSUMERS
Internal Appsweb · service
Agents (3rd-party)OpenAI · Anthropic
Plabs Orchestratorsprebuilt UIs
Workforceanalyst · ops
External Partnersscoped
CLI / Notebooksdata sci
SURFACE
OpenAPI · REST tool surfacetyped · versioned · documented
MCP · agent-native tool surfacediscovery · streaming
Tool Discovery + AutoCrosswalkemergent composition
CONTROL
AuthZscoped · gated
Auditfull lineage
Rate / Quotaper principal
Hotpath Detectself-streamlining
Self-Healingretry · degrade
ObservabilityOTel · traces
SOURCES
Data WarehousesSnowflake · BQ
Operational DBsPostgres · SQL
SaaS SystemsSalesforce · Workday
Document StoresSharePoint · Confluence
Legacy Mainframevia adapter
Vector / SearchOpenSearch · pgvector
Every call routed through the Fabric carries identity, scope, lineage, and a reversible audit trail. Tools are typed at their boundary; access is governed by policy, not by who happens to know the connection string.
FIG. 0102 / Pillars
What it gives you
Four properties enterprise AI infrastructure has to have.
Governed tool surface
Sources become typed tools published over OpenAPI and MCP. Discoverable, versioned, callable by humans, services, and agents alike.
Self-healing & self-streamlining
The tool suite watches its own usage, retires what no one calls, deepens what runs hot, and retries through transient failure without a person in the loop.
AuthZ, scoping, and audit
Every call carries an identity, a scope, and a trace. Security and compliance get the audit trail they need; teams get the access they need; nobody gets both keys.
03 / Capabilities
Inside the platform
Built for the workloads that broke your last AI project.
CAP / 01DISCOVERY
Data source discovery
Workshops + automated crawling identify every system worth exposing. We score sources for quality, freshness, sensitivity and AI utility, and produce a defensible inventory of what to integrate, in what order, and what to leave alone.
- Typical scope
- 40–200 sources
- Time to map
- 2–4 weeks
CAP / 02SURFACE
OpenAPI + MCP, by default
Every tool is published twice: once as a typed REST surface for services, once as a Model Context Protocol surface for agents. Same governance, same audit, same versioning. No second integration.
- Surfaces
- OpenAPI 3.1 · MCP
- Discovery
- Auto-generated
CAP / 03ORCHESTRATION
Prebuilt orchestrators & UIs
Ready-to-deploy agent and workflow patterns (research analyst, case investigator, ops console) that exercise the fabric out of the box. Or bring your own agents; the fabric is the substrate, not the prison.
- Patterns shipped
- 6+
- BYO agents
- First-class
CAP / 04GOVERNANCE
AuthZ, scoping, audit
Identity flows end-to-end. Tools accept scoped principals; the fabric enforces row-, column-, and operation-level rules; every call is auditable to a person, a session, and an intent. Compliance signs the design once.
- Identity
- OIDC · SAML
- Audit
- Append-only
CAP / 05SELF-HEALING
Hotpath detection
The fabric watches its own traffic. Hot tools get cached, parallelized, and promoted. Cold tools age out. Failure modes self-document. The platform you operate in year three is faster than the one you launched.
- Telemetry
- OTel native
- Mode
- Continuous
CAP / 06AUTOCROSSWALK
AutoCrosswalk emergent tools
When agents repeatedly compose the same tools in the same shape, the fabric proposes a crosswalk: a single high-performance tool that captures the pattern. Deployed, governed, and audited like any other tool. Your platform learns its own shortcuts.
- Detection
- Unsupervised
- Approval
- Human-in-the-loop
Feature spotlight
AutoCrosswalk: the tools your platform discovers it needs.
Agents and humans compose tools in patterns. AutoCrosswalk watches those compositions across the fleet, scores them for frequency and latency cost, and proposes a consolidated tool. Already typed, already governed, already auditable.
Your team approves it. The fabric publishes it. The next time the pattern is requested, one call replaces six. Year-one tools become year-three primitives.
FIG. 02 / CROSSWALK EMERGENCE
04 / Engagement
How we work
A staged engagement, not a procurement spiral.
Discover
Source inventory, scope mapping, governance posture. We arrive at a defensible plan before recommending a single integration.
~ 4 weeks
Stand up
Control plane deployed, first 5–10 governed tools live, OpenAPI and MCP surfaces published, audit running.
~ 6–10 weeks
Operate
Orchestrators integrated, hotpath detection enabled, AutoCrosswalk monitoring traffic. Internal teams onboarded.
~ 6–12 weeks
Hand off
Your team operates the fabric. Plabs stays as strategic advisor: quarterly reviews, new-tool patterns, escalation.
Ongoing
05 / Specifications
For the architects
The honest answers to your honest questions.
- Deployment
- Runs in your cloud, on your accounts. We never centralize your data. Deploys to AWS, GCP, or Azure; on-prem available with enterprise terms.
- Identity
- Federates with your IdP via OIDC or SAML. Tool calls carry user identity end-to-end; service-to-service calls carry workload identity. No shared secrets in flight.
- Audit
- Every tool invocation is logged with principal, scope, parameters, response shape, and trace ID. Append-only. Exportable to your SIEM.
- Models
- Model-agnostic. We support Anthropic, OpenAI, Google, AWS Bedrock, and self-hosted models. Tool routing decisions are a configuration, not a rebuild.
- Data residency
- Fabric components run in your region. Configurable per-source. PHI/PCI workloads on dedicated tenancy with documented controls.
- Pricing
- Engagement-based, not per-seat. Discovery + stand-up + operate is a fixed scope; ongoing advisory is separately contracted. We do not mark up infrastructure.