01 Background
Source checks
Verify repository, package hash, dependencies, model provider, runtime, owner, declared tools, and data classes before access is granted.
Product
The agent firewall control plane.
Anubis gives security, platform, and AI teams one place to verify agent source, validate behavior, approve access, track runtime activity, manage risk, and revoke trust.
02 Validation
Safety testing
Evaluate prompt injection, PII exposure, unsafe tool calls, policy bypass, and permission fit against the agent's declared purpose.
03 Access
Controlled release
Approve scopes, issue short-lived permissions, enforce policy at runtime, and remove access when risk changes.
Control model
Every agent becomes a security object.
Anubis models agents as governed identities with source, purpose, safety results, permissions, runtime traces, owners, and evidence attached.
agent firewall objectfinance-reconciler
Layer
Source
Safety
Access
Evidence
Identity
repo signed
owner mapped
review required
passport
Tools
MCP manifest
tool misuse tested
short token
call log
Data
declared classes
PII checks
residency policy
export
Runtime
langgraph
policy guard
block unsafe
trace
Management
Manage the full agent security loop.
01
Inventory
Maintain a living registry of internal agents, vendor agents, MCP-enabled workflows, and shadow deployments.
02
Validate
Run background checks and safety tests against real access requests before production deployment.
03
Approve
Route sensitive access through the right business, security, and data owners.
04
Guard
Monitor behavior, block drift, retain evidence, and revoke trust across connected systems.