inhibitor-lab

inhibitor-lab is the open-source R&D space for integrating Inhibitor Service 3.1.10 into agent-based systems. The audience is AI technologists who need defensible, real-time controls to keep autonomous agents from doing unsafe or wasteful work.

---

Philly Codefest OpenBuild kit

• Start here: codefest/README.md
• Team operations: codefest/TEAM_MANAGEMENT.md

---

Why teams ship with the Inhibitor

• Real-time inhibition for LLM-driven agents, with interruptibility and course correction rather than post-hoc filters.
• Audit-ready trails that explain why actions were blocked, enabling compliance reviews and root-cause analysis.
• Deployment agility via edge-first, stateless design that respects geographic data boundaries and GDPR expectations.

Industry signal: AI Safety Index (July 2025)

Leading AI companies scored D or worse on existential safety and lacked real-time decision safeguards. The Inhibitor is designed to close that gap with inline ethical reasoning and actionability.

References:

• Future of Life Institute – Full Report
• Local PDF Copy

---

Core capabilities

• Insight Mode – slower, with narrative reasoning for compliance, audits, and debugging.
• Performance Mode – fast, flag-only moderation for real-time agents and high-throughput tasks.
• GDPR compliance by design – data minimization, single-purpose processing, and stateless operation. See docs/gdpr-compliance.md.
• Global edge deployment – compute stays close to users to reduce cross-border movement. See docs/global-edge-deployment.md.
• Explained oversight – see docs/inhibitor-inside.md for a narrative on how the Inhibitor converts agent thoughts into structured, ethical decisions.
• See it live – explore the Inhibitor Demo to watch how risky prompts are detected, interventions fire, and responses get redirected in real time.

---

Quickstart (developers)

1. Get an API key from appliedAIstudio.
2. Review the REST API: docs/inhibitor-api.md.
3. Fetch the latest OpenAPI spec (no API key required):
   • JSON: curl https://iaas.appliedai.studio/openapi.json
   • YAML: curl https://iaas.appliedai.studio/openapi.yaml
4. Run an example notebook (pick one):
   • Quickstart: Inhibitor API
   • Adaptive Feedback Agent
   • Real-Time Moderation Agent
   • Indirect Injection Boundary Matrix
   • Legitimate-Use PII Evaluation
5. Layer the Inhibitor into your agent loop (oversight, critique, and action correction).
6. Stress-test and monitor using the Inhibitor Application Sprint.

---

Compliance and regulation coverage

The Inhibitor tracks and surfaces violations across a maintained set of regulations. See docs/supported-regulations.md for the latest mappings.

---

Repository map

• notebooks/ – interactive scenarios showing oversight and moderation patterns.
• examples/ – lightweight code snippets for quick integration.
• docs/ – API reference, deployment guidance, and governance resources.
• benchmarks/ – entry point for supported workflows and historical benchmark assets.
  • benchmarks/supported/ – actively maintained benchmark workflows.
  • benchmarks/supported/operational/ – supported operational load and recovery benchmark.
  • benchmarks/supported/runtime_trajectories/ – supported controlled runtime trajectory benchmark.
  • benchmarks/legacy/ – historical benchmark implementations and result artifacts.
• codefest/ – Philly Codefest OpenBuild challenge kits, starter notebooks, shared logs, and team operations docs.

Benchmarks

Start with the benchmark lifecycle guide. Supported benchmarks are the authoritative entry points for new execution, while historical benchmark implementations and outputs remain available under benchmarks/legacy/ for provenance and comparison.

• Supported operational benchmark – review the operational guide, run the benchmark notebook, and use benchmark_reporting.py for its deterministic reporting workflow.
• Supported controlled runtime trajectory benchmark – see the runtime trajectory guide and use the runner dry-run to list and validate its controlled scenarios.
• Operational reference report – see the published Inhibitor 2.24.1 reference run in benchmark_report.md.

Documentation trails

• Beyond prompt screening – Explore broader uses for Inhibitor, including checks around tool actions, data writes, outbound responses, multi-step behavior, and human review.
• Real-time applications – Use the real-time application best practices to design bounded checks, safe fallbacks, and deterministic enforcement, with a voice agent as the working example.
• Implementation flow – Start with the Inhibitor Application Sprint, then dive into the Reason-Observe-Adjust pattern and the Inside the Inhibitor narrative to see how the oversight loop runs.
• Adoption progression – Follow the typical Inhibitor adoption progression to sequence confidence calibration, schema hardening, error prevention, and auditability.
• Case studies – See how the inhibitor performs in production with the healthcare case impact report.
• Policy to runtime – Trace how written policies become DILL rules in the policy-to-rule examples, then compare enforcement outcomes in the API reference and supported regulations.
• Governance and compliance – Pair GDPR guidance with global edge deployment and zoom out to the ethical inference theory that underpins the system.
• Change history – Start with the Inhibitor Service 3.1.10 release notes for the latest agent intent guidance, then review release notes 3.1.9 for earlier real-time context guidance.

---

Build with confidence

• Walk the full implementation playbook: Inhibitor Application Sprint.
• Need help or an enterprise key? Visit appliedAIstudio.