Sanctum Runtime
Sanctum Runtime is middleware that checks actions from AI agents, robotics, automation, and other systems before they affect the real world.
What Sanctum Runtime is
Sanctum Runtime is middleware that checks actions from AI agents, robotics, automation, and other systems before they affect the real world. It evaluates actions against policies and can approve, require verification, or block them. Operators can review, approve, deny, or escalate actions through web and mobile workflows. The product describes behavioral monitoring, audit logs, policy history, and decision trails for visibility and incident review. Its open-core runtime supports local policy enforcement and offline checks, including local Ollama risk scoring. Documented integrations include OpenAI, Claude/MCP, Gemini-style function calling, DeepSeek, Grok, Bedrock-compatible endpoints, NVIDIA NIM, custom agents, and local runtimes. Developers have Node and Python SDKs, an MCP wrapper, ROS2-oriented policy packs, and smart-home and embodied-agent templates. Security mechanisms include signed agent and action tokens, short-lived approvals, source-trust context, blast-radius scoring, audit-chain verification, and encrypted platform keys. Sanctum is self-hostable under an MIT license and also offers a hosted console. It describes itself as embedded middleware, not a standalone consumer app or robot controller.
Good
- Policies can approve, verify, or block actions
- Local policy checks support offline operation
- Web and mobile operator review workflows
- Self-hostable under an MIT license
- Node and Python SDKs are listed
Watch out
- It is middleware, not a standalone consumer app
- It is not a robot controller
- Some enterprise capabilities are private
Who it suits
Sanctum Runtime is for teams building AI agents, robotics, workflow automation, or systems that take real-world actions. Its stated audiences include agent startups, platform and AI engineers, and security or compliance teams.
Verdict
Sanctum Runtime provides policy checks, operator approvals, and audit visibility for teams building systems that can act in the world. It is embedded middleware, so it is not a standalone consumer app or robot controller.
The facts
Each one read from Sanctum Runtime's own pages, with the date it was checked.
What it does
- Action controls
- It evaluates actions against policies and can approve, require verification, or block them before execution.checked 4 Oct 2026
- Action policies
- Policies can approve, require verification for, or block actions before execution.checked 9 Oct 2026
- Audit
- Runtime audit trails can include action requests, decisions, policy evaluations, approval states, anomalies, and timestamps.checked 9 Oct 2026
- Developer access
- Developers can install the Node SDK and use an agent adapter, an MCP wrapper, or a hosted OpenAI-compatible proxy.checked 4 Oct 2026
- Developer support
- The documentation lists Node and Python SDKs, an MCP wrapper, ROS2-oriented policy packs, and smart-home and embodied-agent templates.checked 9 Oct 2026
- Enterprise capabilities
- The documentation lists fleet orchestration, organization-wide policy sync, advanced analytics, hosted cloud, compliance, and device attestation as private enterprise features.checked 9 Oct 2026
- Hosted access
- The documentation describes a hosted proxy path for OpenAI-compatible traffic and a cloud dashboard for operator workflows.checked 9 Oct 2026
- Integrations
- The same verification path is described for OpenAI, Claude/MCP, Gemini-style function calling, DeepSeek, Grok, Bedrock-compatible endpoints, NVIDIA NIM, custom agents, and local runtimes.checked 9 Oct 2026
- Intended users
- The site presents the product for agent startups, platform and AI engineers, and security or compliance teams.checked 4 Oct 2026
- License and deployment
- The repository is described as MIT-licensed and self-hostable; the product also offers a hosted console, while some enterprise capabilities are private.checked 4 Oct 2026
- Limits
- Sanctum describes itself as embedded runtime middleware, not a standalone consumer app or robot controller.checked 4 Oct 2026
- Local operation
- The runtime supports local policy enforcement and offline verification.checked 9 Oct 2026
- Monitoring
- Sanctum describes behavioral monitoring for unusual command sequences, escalation patterns, suspicious remote instructions, and policy violations.checked 9 Oct 2026
- Monitoring and audit
- Sanctum describes behavioral monitoring, audit logs, policy history, and decision trails for runtime visibility and incident review.checked 4 Oct 2026
- Offline operation
- The open-core runtime supports local policy checks and offline heuristics, including local Ollama risk scoring.checked 4 Oct 2026
- Open source
- The documentation says the repository is MIT-licensed and describes the product as open-core.checked 9 Oct 2026
- Purpose
- Sanctum is a runtime trust layer that evaluates AI actions before they execute.checked 9 Oct 2026
- Robotics support
- The documentation lists Python SDK support, ROS2-oriented policy packs, and smart-home and embodied-agent templates.checked 4 Oct 2026
- Security mechanisms
- The documented security layer includes signed agent and action tokens, short-lived approvals, source-trust context, blast-radius scoring, audit-chain verification, and encrypted platform keys.checked 9 Oct 2026
- Who it serves
- The site describes the product for teams building AI agents, robotics, workflow automation, and other systems that can take real-world actions.checked 9 Oct 2026
Where it runs
- Human approvals
- Operators can review, approve, deny, or escalate actions through web and mobile review workflows.checked 4 Oct 2026
Alternatives to Sanctum Runtime
- 1
PromptGuard
Free planWebAPIExtensionLinux7.9From $19 a month - 2
Prompt Inspector
Free planOpen sourceWebAPILinuxmacOS7.6Free plan - 3
CyberSecEval
Free planOpen sourceWebAPILinuxmacOS7.5Free plan - 4
NVIDIA NeMo Guardrails
Free planOpen sourceAPILinuxmacOSSelf-hosted7.4Free plan - 5
Openlayer Guardrails
Free planWebAPISelf-hosted6.9Free plan - 6
GuardionAI
WebAPILinuxmacOS6.8