Deterministic policy gates
Evaluate every attempted tool call against declarative capability policy before execution.
ops-agent-1 · capfence.enforce()capfence check ./agentsscore 0.03PASSpayment.transfer amount=5000score 0.42REVIEWrm -rf /var/lib/postgresqlscore 0.94BLOCKmcp.read ../../secretsscore 0.87BLOCKDeterministic. Fail-closed. Local-first.
Authorize agent execution before it reaches production.
CapFence sits between AI agents and privileged targets, enforcing capability policy, approval gates, and tamper-evident audit trails before actions reach APIs, databases, filesystems, shells, or MCP servers.
Decision latency<1ms
Default postureFail closed
Audit modelHash chained
Runtime modeLocal first
01 - Platform
Enterprise authorization primitives for agentic execution.
Human approval grants
Queue high-risk actions until an expiring, scoped approval is issued for the exact capability.
Tamper-evident audit
Commit every allow, deny, and approval decision into a verifiable hash chain for incident replay.
Agent framework adapters
Wrap OpenAI Agents SDK, LangChain, LangGraph, CrewAI, AutoGen, MCP, and custom tool runtimes.
02 - Coverage
Built for the places agent mistakes become incidents.
Use CapFence as a local-first control plane for production shell access, payment workflows, database operations, MCP tools, and multi-agent handoffs.
DevOps shell control
Block destructive commands, production mutations, and privileged automation drift.
Database write gating
Deny DDL/DML and high-risk analytics queries before they hit production pools.
MCP boundary security
Proxy tool calls and filesystem access through deterministic host and workspace policies.
Trust lineage
Track multi-agent provenance and prevent unverified nodes from reaching privileged tools.
03 - Docs
Formatted from the CapFence repository docs.
04 - Cases
Blueprints for high-risk autonomous workflows.
05 - Blog
Notes on agent security, DevOps controls, and supply-chain trust.
AI Security
Agentic AI Security Is Moving From Prompts to Runtime Control
Why tool misuse, identity abuse, MCP expansion, and approval gaps are making execution authorization a core AI security control.
DevOps SecurityLeast Agency for Enterprise DevOps
Agentic DevOps needs the same discipline as production access: scoped authority, approval paths, and audit trails.
Reach out
Building autonomous systems with real operational authority?
Talk to the owner about CapFence integrations, enterprise use cases, or early deployments.