|
EvalView – pytest for AI Agents | Open Source Agent Testing Framework
EvalView is an open-source testing framework that brings pytest-style workflows to AI agents. Generate tests, track regressions, and keep tool-calling agents...
|
82 % |
|
Vizra Evals — Agent Evals for the Laravel AI SDK, Written in Pest
Built for the official Laravel AI SDK: write agent evals as Pest tests, record every run — scores, judge reasoning, cost — and fail CI on regressions instead...
|
82 % |
|
Inspect - Turn ideas into mergeable pull requests
Inspect turns ideas into mergeable pull requests. It works with the same context a real engineer has—codebase, dependencies, database, CI, logs, errors, anal...
|
81 % |
|
Code checks for problems linters miss | Scruple
Catch valid code that breaks your team's standards with named, tested checks that run on every change.
|
80 % |
|
Frontguard — Catch the regression, not the noise
AI-powered visual regression testing. AI vision tells a real regression from an intentional change or content, so a red run means something again. Open-sourc...
|
80 % |
|
AI Agent Security Scanner — Vibefixing
Scan your AI agent repo and find unsafe tool calls before shipping. Detect prompt injection, unguarded Stripe charges, and DB mutations. Free scan.
|
80 % |
|
Evals Weekly
The useful ideas in AI testing, explained in five minutes. Research, company posts, and practical steps you can try.
|
80 % |
|
promptry · Sentry for prompts
Local-first prompt observability for LLM pipelines. Versions prompts, runs eval suites in CI, tracks per-call cost with budgets, captures traces and feedback...
|
80 % |
|
Mostafa Mamdoh (WaltWhitehat) — Application Security Researcher
WaltWhitehat is Mostafa Mamdoh, independent application security researcher: web and API penetration testing, specialising in authorization and payment logic.
|
80 % |
|
Wystan — Product-aware change intelligence for software teams
Wystan maps code changes to named product features and user flows, then brings structural reach, evidence-backed watchlists and runtime signals into pull req...
|
80 % |
|
Seer - Stop guessing on retrieval quality
Seer tracks groundedness, recall, and latency on every agent interaction so you catch drift before customers do.
|
79 % |
|
crimes: change risk for TypeScript, JavaScript and Python
Open-source CLI that scans TypeScript, JavaScript & Python repos for change risk and AI-agent risk, including disagreements between languages. Runs locally a...
|
79 % |
|
RedCrown | AI regression checks and release evidence
Check AI changes against defined requirements, inspect failures, and review the evidence behind the result. Start with a supported example or your own outputs.
|
79 % |
|
Agent Jig — Test your AI agent before your users do.
Evaluation and testing framework for AI agents. Define test scenarios in YAML, run against your agent in CI, get pass/fail scores and regression diffs.
|
79 % |
|
ConeReplay — Pre-deploy regression testing for AI agents
Pre-deploy regression testing for multi-agent AI. Replay prompt, model, and agent changes against recorded production traces. Patent pending.
|
79 % |
|
Isolate — Pinpoint which component improved your AI agent
Isolate tells AI developers which part of their agent is working. Run controlled experiments to attribute performance changes to specific components.
|
79 % |
|
Larity — Prompt evaluation for teams
Build, test, and compare prompts against real datasets with structured evaluations and iteration history.
|
79 % |
|
ReliAgent — know if your agent's last move actually worked
ReliAgent checks every AI agent tool call for hallucination, repetition loops, schema violations, confidence collapse, and four more failure modes — one POST...
|
79 % |
|
spanview — read the verdict, not the logs
spanview ingests OpenTelemetry traces from any AI agent, runs an LLM evaluator over every step, and tells you exactly what went wrong — and what changed sinc...
|
79 % |
|
flamelens — upload a profile, leave with a plan
Upload a JFR profiler recording and get an AI performance report: the bottlenecks, the evidence, and the concrete fixes. Free tier, terminal included.
|
79 % |
Detected social media networks