Skip to main content

Argus Sentry

The Argus boundary for AI agents: one install that puts every tool call, skill, MCP action, and agent-to-agent request of any model through Argus, and makes executors refuse any action that lacks an Argus-signed capability token.

The four capabilities

  1. Enforcement - every agent action evaluated against the operator's policy before it executes, with typed reason codes for every block.
  2. Evidence - every decision signed into a tamper-evident Merkle + Ed25519 chain, verifiable offline with argus evidence verify.
  3. Abstention - graded authority (full, scaled, safe-halt) instead of binary allow or deny, and INCONCLUSIVE fail-closed instead of a guess.
  4. The Record - a signed deliberation record per action: what the agent proposed, what the gate saw, and what happened.

Install

pip install argus-sentry

Requires the Argus SDK v0.8 line (argus>=0.8) and Python 3.10 or newer.

Quickstart

Gate an MCP server in one command:

python -m argus_sentry.mcp.gateway \
  --server-id filesystem \
  --policy policy.txt \
  --policy-signer platform_policy_key_1 --policy-key <hex public key> \
  --class-map class_map.json \
  -- npx -y @modelcontextprotocol/server-filesystem /workspace

Wrap the server with token verification:

python -m argus_sentry.mcp.middleware \
  --server-id filesystem \
  --key-list key_list.bin \
  -- npx -y @modelcontextprotocol/server-filesystem /workspace

Model middleware:

from argus_sentry.adapters.model import ModelGate, ModelToolCall

gate = ModelGate(core)  # core: a SentryCore with prefix "model"
outcome = gate.evaluate(ModelToolCall("openai", "delete_all_users", {}, "call-1"))
# outcome.decision == "deny", with the typed reason; nothing executes.

Evidence chain:

from argus_sentry.evidence import SentryChainRecorder

chain = SentryChainRecorder(record_key=record_key, load_authority_key=load_key)
chain.start(policy_hash, policy_source=text)
# wire as the SentryCore recorder: every decision is signed into the chain
chain.export("gate_chain.json")
# then, offline: argus evidence verify gate_chain.json --level L2

Contents

Area What ships
Token library Capability tokens: issue, verify, revoke, replay protection; Python and TypeScript byte-identical to conformance vectors.
MCP adapter Stdio gateway and drop-in server middleware with tool definition pinning.
Model adapters OpenAI, Anthropic, and Google tool-calling middleware, one shared engine, identical decisions.
Framework hooks LangGraph, OpenAI Agents SDK, CrewAI, AutoGen, and A2A.
Executor guards Shell, file, database, and HTTP guards with incident replays.
Evidence chain SentryChainRecorder with the third-party offline verification walkthrough.
Policy packs Coding, customer support, browser, data, and robotics packs mapped to OWASP ASI01 to ASI10 and NIST agent identity controls.

License: proprietary. See the repository for the specification, the conformance matrix, and the verification walkthroughs.

Metadata

Release files for argus-sentry 1.0.1

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

Built distribution (wheel)

Table of built distributions (wheels) for argus-sentry 1.0.1
File Interpreter ABI Platform Actions
argus_sentry-1.0.1-py3-none-any.whl Python 3 none any Details for argus_sentry-1.0.1-py3-none-any.whl

Release files / argus_sentry-1.0.1-py3-none-any.whl

Download URL argus_sentry-1.0.1-py3-none-any.whl
Size 249.2 kB
Tags Python 3
SHA-256 checksum
How to use checksums
3bca169fc02a1484cb9b10ab6b3b5c3d89bc0bb4c7ae023da5f7856f527a85ab
BLAKE2b-256 checksum
How to use checksums
af6a50363317c4e0a86e059168cf2060ce872913c433d28966304d4bd2dfc0db
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/6.1.0 CPython/3.11.17

Security for argus-sentry 1.0.1

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.

No provenance found

None of the files in this release were uploaded with provenance metadata.

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

1.0.1 This release

1 release file No 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