Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

Sanctum Runtime

sanctumruntime.com

  • Free plan
  • Linux
  • Web
  • iPhone and iPad
  • Android
  • API
  • Self-hosted

What Sanctum Runtime is

Sanctum Runtime is a trust layer that checks actions from AI agents and other systems before they are carried out. It evaluates proposed actions against policies and can approve them, require verification or block them. Operators can review, approve, deny or escalate actions through web and mobile workflows. Monitoring features include behavioral monitoring, audit logs, policy history and decision trails for incident review. The open-core runtime supports local policy enforcement and offline verification, including local Ollama risk scoring. Developers can use a Node or Python SDK, an agent adapter, an MCP wrapper or a hosted OpenAI-compatible proxy. Documented integrations include OpenAI, Claude/MCP, Gemini-style function calling, DeepSeek, Grok, Bedrock-compatible endpoints, NVIDIA NIM, custom agents and local runtimes. Robotics materials include 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 MIT-licensed and self-hostable, with a hosted console also available. It is embedded middleware, not a standalone consumer app or robot controller; some enterprise capabilities are private.

Who it suits

It is intended for teams building AI agents, robotics, workflow automation and other systems that can take real-world actions. Agent startups, platform and AI engineers, and security or compliance teams are among the audiences it names.

Good to know, and what to watch

  • Can approve, verify or block actions before execution.
  • Supports local policy enforcement and offline verification.
  • Provides audit trails and policy history.
  • Includes Node and Python SDK support.
  • Can be self-hosted under an MIT license.
  • It is middleware, not a standalone consumer app.
  • It is not a robot controller.
  • Some enterprise capabilities are private.

The verdict

Sanctum Runtime provides policy checks and operator review for systems that can take real-world actions. Teams should account for its embedded-middleware role and the private status of some enterprise features.

What the maker publishes

Action controls
It evaluates actions against policies and can approve, require verification, or block them before execution. source
Action policies
Policies can approve, require verification for, or block actions before execution. source
Audit
Runtime audit trails can include action requests, decisions, policy evaluations, approval states, anomalies, and timestamps. source
Developer access
Developers can install the Node SDK and use an agent adapter, an MCP wrapper, or a hosted OpenAI-compatible proxy. source
Developer support
The documentation lists Node and Python SDKs, an MCP wrapper, ROS2-oriented policy packs, and smart-home and embodied-agent templates. source
Enterprise capabilities
The documentation lists fleet orchestration, organization-wide policy sync, advanced analytics, hosted cloud, compliance, and device attestation as private enterprise features. source
Hosted access
The documentation describes a hosted proxy path for OpenAI-compatible traffic and a cloud dashboard for operator workflows. source
Human approvals
Operators can review, approve, deny, or escalate actions through web and mobile review workflows. source
Integrations
The documented verify path supports OpenAI, Claude/MCP, Gemini-style function calling, DeepSeek, Grok, Bedrock-compatible endpoints, NVIDIA NIM, custom agents, and local runtimes. source
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. source
Intended users
The site presents the product for agent startups, platform and AI engineers, and security or compliance teams. source
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. source
Limits
Sanctum describes itself as embedded runtime middleware, not a standalone consumer app or robot controller. source
Local operation
The runtime supports local policy enforcement and offline verification. source
Monitoring
Sanctum describes behavioral monitoring for unusual command sequences, escalation patterns, suspicious remote instructions, and policy violations. source
Monitoring and audit
Sanctum describes behavioral monitoring, audit logs, policy history, and decision trails for runtime visibility and incident review. source
Offline operation
The open-core runtime supports local policy checks and offline heuristics, including local Ollama risk scoring. source
Open source
The documentation says the repository is MIT-licensed and describes the product as open-core. source
Purpose
Sanctum is a runtime trust layer between AI reasoning and real-world execution for agents, robotics, automation, and embodied systems. source
Purpose
Sanctum is a runtime trust layer that evaluates AI actions before they execute. source
Robotics support
The documentation lists Python SDK support, ROS2-oriented policy packs, and smart-home and embodied-agent templates. source
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. source
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. source
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. source

Checked against the maker's pages on 9 Oct 2026.