Every AI system, known, governed and provable.
Custos finds the AI your organization actually uses, applies your policies to every prompt, response and agent action, and keeps the evidence ready for auditors, regulators and records requests.
For government agencies and regulated enterprises. Built for Microsoft 365, Purview and Entra ID environments.
You cannot govern what you cannot see.
Custos scans browsers, endpoints, SaaS, cloud accounts and API traffic continuously. Every AI tool, model, agent and API key it finds is attributed to a person and a department, then placed in one registry.
Shadow AI, named
Generative AI in the browser, plugins inside approved SaaS, unsanctioned API keys in code. Each finding shows who used it and what data it could reach.
One registry of record
Applications, models, agents, vendors, datasets and use cases, each with a business owner, technical owner, risk tier, approval status and renewal date.
Relationships, mapped
See which agents call which models, which models touch which datasets, and which vendors sit behind them. Retire one asset and know what breaks.
Ten modules. One governance record.
Every module writes to the same registry, the same control set and the same audit trail. Start with discovery and add the rest as your program matures.
AI and shadow AI discovery
Continuous scanning across browsers, endpoints, SaaS, cloud and APIs, with every finding attributed to a user and department.
Used by Security, IT
- Shadow AI detection
- SaaS and generative AI discovery
- Model, agent and API discovery
- Browser and endpoint visibility
- Cloud AI service discovery
- Plugin and tool discovery
- API key and credential discovery
- New AI service alerts
- User and department attribution
- Data-source discovery
Systems, models, agents and vendors
A single registry of record for every AI asset, with owners, purpose, data classification, lifecycle and dependencies.
Used by AI office, IT
- Application, model and agent registries
- Use-case and dataset registries
- Vendor and API registries
- Business and technical owners
- Data classification
- Integration and dependency mapping
- Environment and version tracking
- Approval and compliance status
- Review and renewal dates
- Relationship visualization
Policies, approvals and responsible AI
Versioned policies, approved and blocked catalogs, and risk-based approval workflows that route each request to the right reviewers.
Used by AI office, legal, privacy
- Central AI policy library
- Approved, restricted and blocked catalogs
- Role, department and geographic policies
- Employee request portal
- Automated intake questionnaires
- Security, privacy, legal and records review
- Conditional approval and expiry
- Periodic recertification
- Policy acknowledgements
- Human oversight and impact assessments
AI risk and third-party risk
Configurable scoring for inherent and residual risk across privacy, security, bias, hallucination, agentic and regulatory dimensions.
Used by Risk, procurement
- Automated risk assessments
- Inherent and residual scoring
- Configurable methodology
- Risk heat maps
- Bias, fairness and hallucination risk
- Agentic AI risk
- Records and data-residency risk
- Treatment plans and acceptance
- Vendor questionnaires and scoring
- Subprocessor and certification tracking
Controls, frameworks and evidence
A common control set mapped across frameworks, with automated evidence collection, freshness monitoring and audit-ready packages.
Used by Compliance, internal audit
- Framework packs and custom frameworks
- Common control framework
- One-to-many control mapping
- Automated compliance checks
- Control testing and effectiveness
- Gap identification and remediation
- Automated evidence collection
- Evidence expiry and freshness
- Attestations and exceptions
- Auditor evidence packages
Runtime governance and guardrails
Real-time inspection of prompts, outputs, tool calls and agent actions, enforcing policy before data leaves your boundary.
Used by Security, privacy
- Prompt and output monitoring
- Agent and tool-call monitoring
- Real-time policy enforcement
- PII, PHI and PCI detection
- Source code and secret detection
- Prompt injection and jailbreak detection
- Redaction, masking and tokenization
- User coaching before submission
- Model and workflow drift detection
- Kill switch and agent suspension
Audit trails, records and forensics
Tamper-resistant logging of every AI interaction, with retention, legal hold, eDiscovery and full event reconstruction.
Used by Audit, records, legal
- Complete AI activity logging
- Prompt, output and agent logs
- Policy decision logs
- Administrative change history
- Retention schedules and labels
- Legal hold and preservation
- eDiscovery and records search
- Public-records request support
- Chain of custody
- Event reconstruction and timelines
Incident investigation and remediation
Incidents open automatically from violations, leaks and unauthorized AI, with a workspace to investigate, remediate and close.
Used by Security, compliance
- Automatic incident creation
- Manual incident reporting
- Severity classification
- Investigation workspace
- Evidence collection
- Event timeline
- Root-cause analysis
- Corrective actions
- Escalation workflows
- Post-incident review
FinOps, licenses, adoption and ROI
Spend by token, model, agent, vendor and department, with budgets, anomaly alerts, license reclamation and ROI reporting.
Used by Finance, leadership
- AI spend dashboard
- Token and API consumption
- Department and use-case cost
- Budgets and spend alerts
- Cost anomalies and forecasting
- Model cost comparison
- Chargeback and showback
- License utilization
- Adoption by department and tool
- ROI and executive reporting
Microsoft Purview, Entra ID and more
Native Microsoft 365 integration plus an extensible connector and API framework for security, IT, GRC and business systems.
Used by IT, security
- Microsoft Purview and MIP
- Sensitivity and retention labels
- Entra ID, SAML and OIDC
- Microsoft Graph and Copilot
- SharePoint, Teams and Outlook
- SIEM and SOAR
- DLP, CASB and EDR
- ITSM and ticketing
- GRC and records systems
- Webhooks and REST APIs
Policy that acts before the prompt is sent.
Guardrails inspect prompts, files, responses and tool calls in real time. You decide, per department, role or data type, whether Custos allows, coaches, redacts or blocks.
- Social Security numberPII
- Diagnosis and medicationPHI
- Case record from SharePointLabel: Confidential
- Destination toolRestricted catalog
Sensitive data
PII, PHI, payment data, financial records, source code, credentials and confidential documents.
Attacks on the model
Prompt injection, jailbreak attempts and unsafe content, in prompts and in uploaded files.
Outputs too
Responses are checked for leaked data and restricted content before they reach the user.
Redact, mask, tokenize
Keep the work moving without the sensitive values ever leaving your boundary.
Prove a control once. Satisfy every framework.
Custos maps each control to every requirement it satisfies. One piece of evidence, collected automatically and kept fresh, covers the same obligation across all your frameworks.
| Control | NIST AI RMF | ISO/IEC 42001 | ISO/IEC 27001 | NIST CSF | SOC 2 | EU AI Act | GDPR |
|---|---|---|---|---|---|---|---|
| AI system inventory | |||||||
| Risk assessment before deployment | |||||||
| Human oversight of high-risk use | |||||||
| Prompt and output logging | |||||||
| Sensitive data protection | |||||||
| Third-party AI vendor review | |||||||
| Incident response for AI events | |||||||
| Transparency to affected people |
Agents act on their own. Custos sets the limits.
Each agent gets an identity, an owner, a permission set and approval thresholds. Every tool call is logged, drift is flagged, and one switch stops it everywhere.
Registry and identity
Purpose, owner, risk rating, model and tool inventory for every agent.
Human in the loop
Approval points and transaction thresholds for consequential actions.
Behavior and drift
Alerts when an agent calls new tools, new data or a changed model.
Stop instantly
Suspend one agent or all agents of a type, with the action recorded.
Reconstruct any AI event, minute by minute.
Who asked, which model answered, what data was touched, what the policy decided and what happened next. Custos rebuilds the full sequence and preserves it as evidence.
Complete audit trail
Prompts, outputs, model and agent activity, tool calls, data access, approvals and configuration changes, timestamped and attributed. Tamper-resistant where you require it.
Records and eDiscovery
Retention schedules and labels for AI conversations, legal hold, public-records request search, chain of custody and defensible deletion.
Incident response
Automatic incidents from policy violations and leaks, an investigation workspace, root cause, corrective actions and post-incident review.
Know what AI costs, and what it returns.
Token, API, model, agent and license spend, allocated by department, application and use case. Set budgets, catch anomalies and reclaim licenses nobody uses.
Chargeback and showback
Allocate cost to the teams that incur it.
License optimization
Find inactive and underused Copilot and AI seats.
Adoption and ROI
Active users, time saved and value by use case.
Ask your governance program a question.
The built-in assistant answers from your own registry, controls, evidence and logs. Every answer cites its sources, and it only shows what your role is permitted to see.
Four departments had unauthorized generative AI activity in the last 30 days. Human Services accounts for most of it.
| Department | Tools | Users |
|---|---|---|
| Human Services | 2 | 31 |
| Engineering | 3 | 12 |
| Public Works | 1 | 6 |
| Legal | 1 | 2 |
Answer limited to records your role can access. Illustrative data.
One platform for everyone accountable for AI.
Each team works in its own view, on the same record.
Leadership and AI office
Enterprise posture, high-risk systems, spend and adoption in one executive view.
Security
Shadow AI, guardrail events, agent permissions and incidents, fed to your SIEM.
Risk and compliance
Assessments, control testing, evidence freshness and audit-ready packages.
Legal, privacy and records
Impact assessments, retention, legal hold and public-records search.
Finance and procurement
Budgets, chargeback, vendor risk, contracts and license renewals.
Every employee
A simple portal to find approved tools, request new ones, take training and report issues.
Works with the stack you already run.
Deep Microsoft 365 integration for labels, DLP, retention and identity, plus an open connector and API framework for everything else.
Microsoft 365 and Purview
Custos reads your existing labels and policies, so there is nothing to classify twice.
Enterprise ecosystem
SSO through SAML and OIDC, automated user provisioning and deprovisioning, and custom roles with least-privilege access.
Built to hold the most sensitive record you keep.
Custos stores what your people say to AI. It is engineered accordingly.
Encryption
In transit and at rest, with managed keys and secrets handling.
Isolation
Tenant isolation, least-privilege access and privileged-access controls.
Residency
Configurable data residency and retention to meet your obligations.
Resilience
High availability, backup and disaster recovery, with security monitoring throughout.
Developed against ISO/IEC 27001 and SOC 2 criteria, and delivered by SDLC Corp, whose operations are certified to ISO/IEC 27001:2022 and ISO 9001:2015.
Request the security pack- Contact Us
Let's Talk About Your Product
What happens next?
- We review your requirements
- Strategy call with experts
- Clear roadmap & estimate
- NDA Protected
- Enterprise Grade Delivery
- Global Clients