Safety Guardrails and Observability

Kernel architecture for governing autonomous AI agents with policy enforcement.

Diagnostic guardrails that analyze full agent execution trajectories to detect instruction hijacking and tool misuse.

Runtime observability and guardrails for AI agents with loop detection and anomaly alerts.

AI gateway that unifies model providers with automatic failover, load balancing, observability, and MCP support.

Local-first TUI for AI coding agent session observability with tokens, cost, latency, tool failures, anomalies, reports, diffs, and CI health gates.

Pre-action authorization plugin for agent frameworks with policy-based access control.

Hallucination detection system beating GPT-4 baselines, with specialized testing for agent outputs and LLM-generated content quality.

Open-source observability platform built on OpenTelemetry for tracing, evaluating, and debugging AI agents.

Eval-driven development platform with experiment tracking and prompt optimization for agent quality.

Self-hosted observability and opt-in kill switch for coding agents, reading the session logs runtimes already write to disk ([website](https://clawmetry.com)).

Voice agent platform from ElevenLabs for customer support automation with HIPAA compliance and multi-language support.

Semantic memory guardrails using causal graphs to prevent agents from repeating past failures.

Pairs independently configured executors and supervisors with repository-defined quality gates and inspectable evidence for each coding attempt.

Adds structural, type, and quality guarantees to LLM outputs for reliable agent responses.

Open-source LLM observability with one-line integration for cost tracking and prompt analytics.

Real-time protection against prompt injection, data leakage, and toxicity in agent interactions.

Open-source observability and analytics platform purpose-built for the full lifecycle of AI agents.

Open-source LLM observability platform for tracing, prompt versioning, and LLM-as-a-judge evaluations.

LangChain platform for tracing, testing, and evaluating agent performance with production monitoring.

Security toolkit for scanning LLM inputs and outputs to prevent prompt injection and data leaks.

Python-native observability from the Pydantic team with deep integration for high-performance agent monitoring.

Governs autonomous coding agents, turning open-ended runs into budgeted, verified software work with signed outcome receipts.

NVIDIA programmable guardrails toolkit for controlling and securing LLM-powered agent conversations.

Records a coding-agent run below the harness, then replays it offline or forks it onto another model; its MCP server lets an agent read and re-run its own past runs.

Modules for agent runtime security, self-audit trails, and collective cognition patterns.

Security framework covering goal hijacking, tool misuse, and cascading failure mitigations for agents.

Generates cross-tool agent context and privacy-safe evidence receipts for loaded authority, handoffs, and skill use.

Self-hardening prompt injection detection system for securing agent inputs against adversarial attacks.

LLM evaluation framework with 50+ metrics, LLM-as-Judge, and guardrail scanners (jailbreak, PII, injection).

Local, deterministic verification layer for AI agent code: scans, test evidence, and a binding merge verdict without uploading source.

Verifies an agent's outbound requests and MCP handoffs against a source of truth using zero-knowledge proofs, logging each call and blocking anything off the trusted-endpoint allow-list.

Self-hostable end-to-end agent engineering platform with tracing, evals, guardrails, and gateway.

Commit-time audit harness that blocks credential leaks and out-of-scope file changes before they land, with 24 rules and HMAC-chained audit logs.