Clearception
Clearception TrustFabric

One Governance Fabric. Every Agent. Every Action. Every Policy.

Bring existing corporate SDLC, SSDLC, legal, security, compliance and data sovereignty requirements into AI-powered engineering workflows.

Innovate Without Compromise. Govern Without Friction.

The enterprise AI governance challenge

AI coding agents now open pull requests, run tests and touch production pipelines. The policies that govern engineering were not written for them.

  • Policies live in documents; agents live in terminals

    An organization's SDLC standard, secure coding policy and client contracts are written for people. An AI coding agent acting at machine speed never reads them.

  • Every agent has its own permission model

    Claude Code, Codex, Copilot and in-house agents each expose different controls, so the same rule is configured several times, differently, or not at all.

  • Data sovereignty is decided at the model endpoint

    Where a prompt containing client code is processed is a legal question. It is usually answered by whoever wrote the agent's configuration file.

  • Auditors ask what happened, not what was intended

    A policy PDF shows intent. An audit needs a record of each decision, the rule it was made against, and evidence that the record has not been edited since.

How TrustFabric works

People turn policy documents into an approved bundle. A deterministic gateway checks every governed action against it and records the answer.

Policy lane — people decide
  1. Policy sources

    SDLC, SSDLC, security, legal, privacy and sovereignty documents, uploaded or pasted with provenance.

  2. Policy intelligence

    Deterministic extraction of obligation sentences into draft rules mapped to control categories.

  3. Approved bundle

    A person who is not the author approves it. Then it is immutable, versioned and SHA-256 hashed.

The gateway only ever evaluates against an approved bundle version.
Runtime lane — every governed action
  1. Agents and CI

    Claude Code, the Claude Agent SDK, Codex, custom agents and pipeline gate steps send a request.

  2. Authorization gateway

    Evaluates the request against the approved bundle. Missing context or no matching allow rule fails closed.

  3. Decision

    Allow, Deny, Require approval or Insufficient context — with the reasons.

  4. Evidence

    A tamper-evident record in a per-organization hash chain, exportable for audit.

Teal: a human step. Blue: deterministic evaluation, no AI model per decision. Amber: the audit record. Agents can request; they cannot approve their own requests, change policy or mint approvals.

Secure software development

Bring your SDLC and SSDLC standards into the places AI-assisted changes actually happen.

  • Available

    Required checks — branch policy, security scan, SCA, secret detection, IaC validation, SBOM present, tests passed, deployment approval — evaluated against the approved bundle.

  • Available

    Each governed repository declares its classification, so a restricted client repository is treated differently from an internal tool.

  • Available

    Requirement extraction maps obligation sentences in your standards to control categories, as draft rules for people to review.

Intelligent policy enforcement

Policy intelligence proposes; people approve; a deterministic engine enforces.

  • Available

    Policy documents are imported by upload or paste with their source, owner, revision and sensitivity recorded.

  • Available

    Extracted items become draft proposed rules only. No AI model approves anything.

  • Available

    Bundles move draft → in review → approved → superseded. Approval needs an approver who is not the author. Approved bundles are immutable, versioned, diffable and SHA-256 hashed.

  • Planned

    Export of approved bundles as OPA/Rego.

AI agent authorization

Every governed agent action is a request to one gateway, answered against one approved bundle.

  • Available

    Agent inventory: provider, identity, tool permissions, approved repositories and environments, risk class and a kill switch.

  • Available

    An honest enforcement level per agent: monitored (TrustFabric sees reports), voluntary (the agent calls the gateway through a hook it could disable) or enforced (the execution path is mediated).

  • Available

    Enforced today: Claude Agent SDK agents whose tool calls pass through a can_use_tool callback calling the gateway in your own runner, and CI gate steps that fail the job.

  • Planned

    Mediated enforcement for OpenAI Codex and GitHub Copilot. Today they are monitored or voluntary only.

Contract clauses and regulatory obligations become rules the gateway can check, not footnotes in a wiki.

  • Available

    Legal and contractual obligations are extracted, mapped and approved through the same two-person bundle workflow as security policy.

  • Available

    Approval requests and policy exceptions are time-boxed, scoped and decided by a person.

  • Available

    TrustFabric records decisions against your policies. It does not give legal advice and does not by itself make an organization compliant.

Data sovereignty

Decide where code and data may be stored and processed, and which model endpoints may see them.

  • Available

    Admin-defined jurisdictions with allowed storage and processing regions, approved model providers and endpoints, transfer rules by data classification, and retention.

  • Available

    US, EU, UK and Singapore templates are starting points for your own counsel to adjust, not legal advice.

  • Available

    Operations that need a regional execution environment are denied with a reason, because none is provisioned yet. Selecting a cloud region never by itself marks anything compliant.

  • Planned

    TrustFabric-hosted regional execution environments.

CI/CD governance

A pipeline step asks the gateway whether this change may proceed. Anything other than Allow fails the job.

  • Available

    GitHub Actions steps and Azure DevOps script tasks call the gate with a workload token.

  • Available

    Workload tokens (prefix tfw_) are shown once, stored hashed and never hold human-only permissions.

  • In preview

    GitLab uses the same HTTP API; its adapter is in preview.

Continuous audit evidence

Every decision is written down the moment it is made.

  • Available

    Each decision becomes a record in a per-organization HMAC hash chain, with an endpoint that verifies the chain.

  • Available

    Records export as a JSON audit package: the request, the decision, the reasons and the bundle version it was evaluated against.

  • Available

    Records support an audit. They do not establish regulatory compliance on their own, and TrustFabric holds no certifications.

Supported integrations

Where each integration stands today. Planned connectors show as not configured in the product.

  • GitHub

    Available

    Repository connection and Actions gate step.

  • Azure DevOps

    Available

    Pipeline script task calling the gate.

  • GitLab

    In preview

    Same HTTP gate API; adapter in preview.

  • Claude Code

    Available

    PreToolUse hook script calling the gateway (voluntary enforcement).

  • Claude Agent SDK

    Available

    can_use_tool callback in your runner (enforced).

  • OpenAI Codex

    In preview

    Inventory and voluntary gateway calls; no mediated path.

  • GitHub Copilot

    In preview

    Inventory and monitoring only; no mediated path.

  • SharePoint

    Planned

    Policy document connector.

  • Microsoft Purview

    Planned

    Classification connector.

  • ServiceNow GRC

    Planned

    Controls and evidence connector.

  • Azure DevOps wiki

    Planned

    Policy document connector.

  • OPA / Rego

    Planned

    Approved bundle export.

Enterprise architecture

  • Shared multi-tenant service

    Starter, Growth and Business run on shared infrastructure with tenant isolation at the organization boundary.

  • Your agents, your models

    Coding agents keep running where they run today, on your own AI subscriptions. TrustFabric sees the request, not your model traffic.

  • Deterministic decisions

    The gateway is a rule engine evaluated against a fixed bundle version. The same request and bundle give the same answer, and no AI model is called per decision.

  • Human-only authority

    Approving bundles, approving requests and granting exceptions are human permissions. Workload tokens used by agents and CI never hold them.

  • Enterprise SSO through the platform

    Sign-in uses Clearception platform SSO with your enterprise identity provider, on Business and above.

  • Dedicated deployment

    Single-tenant governance infrastructure is an Enterprise or add-on option. It is planned, not yet available.

Pricing

Four plans, priced per organization. Starter, Growth and Business are bought online; Enterprise is quoted.

  • Starter

    $499/month

    10 developers · 10 repositories · 5,000 evaluations/month

  • Growth

    Recommended
    $1,999/month

    50 developers · 50 repositories · 50,000 evaluations/month

  • Business

    $5,999/month

    150 developers · 200 repositories · 250,000 evaluations/month

  • Enterprise

    Starting at$12,500/month

    Capacity negotiated by contract.

Compare plans, add-ons and how billing works

See a governed agent request, end to end

The demo walks one request from a development agent to a restricted client repository through the gateway, with simulated decisions you can change.

Questions

Can I buy TrustFabric today?
Yes. Starter, Growth and Business are sold online, monthly or annually, through Stripe Checkout from the TrustFabric console. Enterprise is quoted — request a demo and we will contact you.
Does TrustFabric include Claude Code, Codex or other AI coding agents?
No. You bring your own AI model and API subscriptions for your coding agents. TrustFabric governs what those agents are allowed to do.
Does an AI model decide whether an action is allowed?
No. Decisions come from a deterministic engine evaluating an approved policy bundle. Requirement extraction from your documents is also deterministic, and its output is draft rules that a person must approve.
What happens when the gateway is missing information?
It fails closed. Missing required context returns Insufficient context, no matching allow rule returns Deny, and an engine error returns Deny.
Can TrustFabric stop GitHub Copilot or Codex directly?
Not today. For those agents TrustFabric is monitored or voluntary: it records what they report or what they choose to ask. Mediated enforcement is available where the execution path passes through the gateway, such as Claude Agent SDK agents in your runner and CI gate steps.
Does TrustFabric make us compliant?
No product does that on its own. TrustFabric records each decision as tamper-evident evidence your auditors can review. It holds no certifications, and its jurisdiction templates are not legal advice.

More in the TrustFabric help center.