Our take
Newless than six months old and changing fast- Activity
- 44
- Community
- 20
- Issues
- 46
- Pull requests
- 63
+0 stars in the last 30 days
- Fewer contributors than 82% of the projects we track
AgentShield is a focused, dependency-light rules engine, but its recent activity and small, concentrated contributor base make it better suited to evaluation and deliberate integration than as an assumed, independently supported hosted service.
Good fit if
- You need Python-based, deterministic transaction checks and can supply the transaction history and rule inputs.
- You want a small standard-library package or MIT-licensed evaluation fixtures to incorporate into your own implementation.
- You are comfortable with the project being consolidated into sipi.bot, with no separate AgentShield hosted offer.
Look elsewhere if
- You need a standalone hosted service or a separate AgentShield distribution program.
- Your stack cannot run Python 3.11 or later, or cannot meet the listed Docker requirement.
- You need a project with broader contributor participation, a release history in the past year, or stronger recent issue closure activity.
All health signals
| Last commit | 2026-08-24 (3 months ago) |
|---|---|
| Commits, last 90 days | 88 |
| Releases, last 12 months | 0 |
| Contributors | 3 (top contributor: 92% of commits) |
| Issues closed, last 90 days | 1 |
| Pull requests merged, last 90 days | 7 (typically merged in 0 days) |
| Project age | 3 months |
Checked on 2026-11-09 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
2Forks
8Change since Nov 15, 2026: 0 stars·0 forks
Overview
AgentShield is a pure-Python spending firewall for AI agents, built on the Python standard library. It addresses the risk of an agent making costly or repeated transactions by evaluating each proposed transaction against spend-control rules before it proceeds.
The package is aimed at developers integrating spending controls into agent workflows. It is stateless and deterministic, and the README says evaluations take under a millisecond. The project has consolidated its spend-control work into sipi.bot; the Python package and test fixtures remain available, but AgentShield no longer has a separate hosted offer or distribution program.
Key Features
- Provides ten composable rule types, including transaction limits, daily totals, velocity checks, and merchant allowlists.
- Supports controls for session budgets, expected cascade costs, human-review thresholds, duplicate transactions, and circuit state.
- Evaluates transactions against supplied rules and prior transactions without file, network, or global-state dependencies.
- Uses
Decimalfor monetary values rather than floating-point arithmetic. - Includes an evaluation suite of 74 scenarios that can also serve as MIT-licensed test fixtures.
- Runs on Python 3.11 or later and has no Python package dependencies, according to the README.
Use Cases
- An AI-agent developer wants to block an individual API charge above a configured limit before authorizing it.
- A team needs daily or per-session spending caps and can provide transaction history to the evaluator.
- A platform wants to restrict transactions to approved merchants or categories.
- A team needs duplicate-transaction checks or a circuit-breaker rule to deny calls when a supplied circuit state is tripped.
- Developers building their own spend-control system want reusable scenarios for testing rule behavior.
What you need
Detected in the repository
- Python >=3.11 (from pyproject.toml)
- A Dockerfile, so it can run in a container
- A test suite
License in plain words
MITpermissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the copyright and license notice
- Share your changes: no
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
pip install agentshield-spend
See the README for usage details and the complete rule reference.
Considerations
- The project is new and changing quickly. It has recent commit activity, but no release in the past year.
- Most contributions come from one person, despite contributions from a small group overall. Few issues have been closed recently, and the available health data has no median issue-resolution time.
- The README describes the evaluator as stateless and expects callers to provide prior transactions for history-dependent rules. It does not describe a built-in persistence or hosted evaluation service.
- Python 3.11 or later is required, and Docker is listed as a requirement. The repository has tests but no CI configured in the supplied project facts.
- The project is MIT-licensed. Its README says the package and fixtures remain available while new implementation work runs through the sipi.bot pilot.
- The README identifies the installation distribution as
agentshield-spendand the import name asagentshield; verify package naming when integrating.
Repository · Canonical product · Paid implementation pilot · Public evaluation report
Found this useful?
Share it with someone who would like agentshield.