Skip to main content

AgentGate

PyPI Python License: MIT

Auth0 for AI agents — verifiable identity, scoped permissions, human-approval escalation.

AgentGate gives every AI agent a cryptographic identity, enforces deny-by-default permissions through YAML-defined policies, and escalates high-risk actions to a human before they execute.

Quick Start

pip install agent-gatekeeper
from agentgate import AgentIdentity, PermissionPolicy, GateKeeper, Action
from agentgate.policy import ActionRule

# 1. Create an agent identity
agent = AgentIdentity.create("research-bot", capabilities=["search", "summarize"])

# 2. Define a permission policy
policy = PermissionPolicy(
    name="research-policy",
    rules=[
        ActionRule(action=Action.TOOL_USE, allowed=True, allowed_targets=["search"]),
        ActionRule(action=Action.LLM_CALL, allowed=True, max_per_minute=10),
        ActionRule(action=Action.SHELL, allowed=False),
    ],
)

# 3. Enforce at runtime
gate = GateKeeper(identity=agent, policy=policy)

decision = gate.check("tool_use", target="search")
print(decision.type)  # DecisionType.ALLOW

decision = gate.check("shell")
print(decision.type)  # DecisionType.DENY

Features

  • HMAC-signed identity — every agent gets a unique key pair for message signing and verification
  • Deny-by-default policies — nothing is allowed unless explicitly permitted
  • YAML round-trip — define policies as code, load and save from YAML files
  • Target-level scoping — allow specific tools, files, or endpoints while denying others
  • Usage thresholds — escalate or hard-deny after N uses of an action
  • Rate limiting — cap actions per minute to prevent runaway agents
  • Human-approval escalation — plug in your own approval handler (Slack bot, CLI prompt, webhook)
  • Decision audit log — full history of every gate check for observability

API Reference

AgentIdentity

Verifiable identity for an AI agent, backed by HMAC-SHA256.

Method Description
AgentIdentity.create(name, capabilities=None) Create a new agent with a generated UUID and secret key
agent.sign(message) -> str Sign a message string, returns hex digest
agent.verify(message, signature) -> bool Verify a signature against this agent's key

Fields: id, name, capabilities, created_at, secret_key, metadata

PermissionPolicy

Deny-by-default permission policy with YAML serialization.

Method Description
policy.check(action, target=None) -> str Returns "allow", "deny", or "escalate"
policy.get_rule(action) -> ActionRule Get the rule for a specific action type
PermissionPolicy.from_yaml(path) Load a policy from a YAML file
policy.to_yaml(path) Save a policy to a YAML file

Actions: tool_use, llm_call, file_read, file_write, network, shell, data_access

GateKeeper

Runtime enforcement engine that combines identity, policy, thresholds, and escalation.

Method Description
gate.check(action, target=None) -> Decision Check permission; returns a Decision object
gate.is_allowed(action, target=None) -> bool Convenience boolean check
gate.require(action, target=None) -> Decision Check and raise PermissionError if denied
gate.decisions -> list[Decision] Full audit history of all gate checks

Decision fields: type (ALLOW/DENY/ESCALATE), agent_id, action, target, reason, timestamp, approved_by

Architecture

┌─────────────────────────────────────────────────────┐
│                    Your Agent                       │
│                                                     │
│   action request                                    │
│        │                                            │
│        v                                            │
│   ┌──────────┐    ┌──────────────────┐              │
│   │GateKeeper│───>│PermissionPolicy  │              │
│   │          │    │  (YAML rules)    │              │
│   │          │    └──────────────────┘              │
│   │          │                                      │
│   │          │──> threshold / rate-limit check      │
│   │          │                                      │
│   │          │──> escalation ──> approval_handler() │
│   │          │                     (human / webhook)│
│   └──────────┘                                      │
│        │                                            │
│        v                                            │
│   Decision(ALLOW / DENY / ESCALATE)                 │
│   + full audit log                                  │
└─────────────────────────────────────────────────────┘
         │
         v
   AgentIdentity
   (HMAC-signed, UUID-backed)

Comparison

Feature AgentGate Raw LLM APIs LangChain Tools Custom middleware
Agent identity with crypto signing Yes No No Manual
Deny-by-default policies Yes No No Manual
YAML policy definition Yes No No No
Per-action rate limiting Yes No No Manual
Human-approval escalation Yes No No Manual
Decision audit log Yes No No Manual
Zero dependencies beyond pydantic Yes N/A Heavy Varies

License

MIT License. See LICENSE for details.

Metadata

Release files for agent-gatekeeper 1.0.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for agent-gatekeeper 1.0.0
File Size Uploaded Provenance Actions
agent_gatekeeper-1.0.0.tar.gz 8.8 kB With provenance Details for agent_gatekeeper-1.0.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for agent-gatekeeper 1.0.0
File Interpreter ABI Platform Actions
agent_gatekeeper-1.0.0-py3-none-any.whl Python 3 none any Details for agent_gatekeeper-1.0.0-py3-none-any.whl

Total release size: 17.7 kB

Release files / agent_gatekeeper-1.0.0.tar.gz

Download URL agent_gatekeeper-1.0.0.tar.gz
Size 8.8 kB
Tags Source
SHA-256 checksum
How to use checksums
767aba356d657f589ba80ad96e834f7a20416d5b44b6a89b9b4a12587666ee9f
BLAKE2b-256 checksum
How to use checksums
57e0e368a494384fee5671d04914511755d2dfbe434a2fc73dff3d288172bcc9
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 4, 2026.

Transparency log

Release files / agent_gatekeeper-1.0.0-py3-none-any.whl

Download URL agent_gatekeeper-1.0.0-py3-none-any.whl
Size 8.9 kB
Tags Python 3
SHA-256 checksum
How to use checksums
05119d332a17682b2586eeedd8614f4c4856c3a2942ac3e0af1dc13494386904
BLAKE2b-256 checksum
How to use checksums
47aacb6c4087f847b12bcd3ef03182ab99ee8a6aa4f17d7072265303b119e050
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.7

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 4, 2026.

Transparency log

Security for agent-gatekeeper 1.0.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for agent_gatekeeper-1.0.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 4, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

1.0.0 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page