Skip to content

Releases: schchit/Agentic-Firewall

Agentic Firewall v3.1: Production-oriented safety and uncertainty control stack for multi-agent systems

08 May 10:49
@schchit schchit
v3.1
5dc4aad
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

Choose a tag to compare

View all tags
Agentic Firewall v3.1: Production-oriented safety and uncertainty control stack for multi-agent systems Latest
Latest

Agentic Firewall v3.1

This release introduces Agentic Firewall + Uncertainty Cascade Engine, a production-oriented safety and cost-control stack for multi-agent systems.

It combines three layers:

  1. A gateway layer for agent-message compression and token-cost reduction.
  2. An Uncertainty Cascade engine for graph-theoretic stability analysis.
  3. An Agentic Firewall for automatic mitigation when cascade risk becomes unstable.

Highlights

Agentic Firewall

  • Added firewall logic for detecting and mitigating unstable agentic states.
  • Supports automatic intervention when uncertainty or cascade risk exceeds safe operating thresholds.
  • Designed for multi-agent workflows where errors, uncertainty, or unstable reasoning can propagate across agents.

Uncertainty Cascade Engine

  • Added graph-theoretic cascade analysis for multi-agent systems.
  • Models how uncertainty can spread across agent interactions.
  • Helps identify unstable states before they create downstream failures.
  • Provides a foundation for safety-aware orchestration and automated mitigation.

Gateway Layer

  • Added gateway interface for transforming agent messages.
  • Supports message compression and token-cost reduction.
  • Designed to sit between agents or orchestration layers as a control point.

API Service

  • Added FastAPI-based service entrypoint.
  • Supports local service execution with Uvicorn.
  • Includes protected /transform endpoint for message transformation.
  • Includes observability endpoints:
GET /health
GET /ready
GET /metrics

API Key Security

  • Added file-based API key support through:
security/api_keys.json
  • Supports scoped access control.
  • Includes least-privilege scopes:
transform
admin
  • /transform requires an API key when security/api_keys.json exists.

Redaction Policy

  • Added tiered redaction support:
redaction_level=standard
redaction_level=strict

This helps reduce sensitive information exposure before messages are routed through downstream systems.

Benchmark and Validation

  • Added benchmark evaluation flow.
  • Supports benchmark gate configuration:
BENCHMARK_GATE=0.95 python benchmarks/evaluate.py
  • Includes unit tests for cascade, firewall, and gateway components:
python -m unittest -v test_cascade.py test_firewall.py test_gateway.py

High-Availability Readiness

  • Added operational validation assets.
  • Includes load smoke tool:
python ops/load_test.py
  • Includes Redis chaos drill playbook:
ops/chaos_redis.md
  • Includes SLO and rollback policy template:
ops/slo.yaml

Docker Support

  • Added Dockerfile and docker-compose support.
  • Start the service with:
docker compose up --build

Quick Start

Install dependencies:

python -m pip install -r requirements.txt

Run tests:

python -m unittest -v test_cascade.py test_firewall.py test_gateway.py

Run benchmark:

BENCHMARK_GATE=0.95 python benchmarks/evaluate.py

Start local service:

uvicorn gateway:app --host 0.0.0.0 --port 8080

Test the transform API:

curl -X POST http://127.0.0.1:8080/transform \
  -H "Content-Type: application/json" \
  -H "x-api-key: sample-transform-key" \
  -d '{"target":"deploy_decision","message":"Fact: stable. Next step: deploy.","strategy":"hybrid"}'

Why this release matters

As multi-agent systems become more complex, failures are no longer limited to a single prompt or model response. Uncertainty, hallucination, overconfidence, unsafe delegation, and poor message routing can cascade across agents.

Agentic Firewall v3.1 introduces a practical control layer for this problem.

It is designed to help agentic systems:

  • reduce unnecessary token cost,
  • compress agent messages into more controllable forms,
  • detect unstable cascade risk,
  • apply automated safety mitigation,
  • expose health, readiness, and metrics endpoints,
  • support API-key-based gateway access control,
  • and provide benchmark-driven validation before production use.

Status

This is an early production-oriented release.

Recommended for:

  • multi-agent safety experiments
  • agentic workflow risk control
  • uncertainty propagation research
  • AI gateway prototypes
  • token-cost reduction testing
  • internal safety and observability evaluation

Not yet recommended as a fully hardened enterprise security layer without additional production review, deployment hardening, monitoring, and policy integration.

License

MIT

Assets 2
Loading