|
gateprobe - approval replay & binding-drift evaluator
Test whether an approval can be reused or bypassed after the target changes. Deterministic, versioned evaluator for agent approval tokens: catches binding dr...
|
80 % |
|
tooldocstest.dev — validate tool quickstart examples offline
Local-only CLI that checks tool quickstart example arguments and runs two allowlisted data-only sandbox tools, emitting deterministic pass/fail/unknown evide...
|
78 % |
|
tooloutbox.dev — durable tool-write outbox
A local-first durable outbox for tool writes: persist an intended write before acknowledgement, and after a crash recover the pending write and reconcile its...
|
78 % |
|
APIDiffGuard: Third-Party API Change Monitoring
Monitor third-party API responses and catch breaking changes like removed fields, type flips and new nulls, even behind a 200 OK. MIT open source, CLI and cl...
|
77 % |
|
Know when your AI gets worse, before your users do | EvalCore
EvalCore is snapshot testing for AI behavior. Record how your LLM app or agent behaves, then replay it in CI on every change: offline, deterministic, $0.
|
77 % |
|
StructTrace | Catch valid-but-wrong structured output regressions
Compare baseline and candidate structured outputs, separate schema validity from task correctness, and generate replayable local release evidence.
|
77 % |
|
Condux: catch every error, conduct every fix
Source-available error monitoring with an AI fix engine that opens draft pull requests your team reviews and merges. Managed cloud or self-host.
|
76 % |
|
Signal Diff — What changed on your site after every deploy
Diff sitemap crawls between deploys: regressions and clears grouped by run, with GitHub Actions and a CLI for AI coding tools.
|
76 % |
|
ClisTa — Effective challenge, on the record.
ClisTa records model risk decisions as accountable state — evidence, the challenges that survived, sign-off authority, tamper-evident provenance — in a seale...
|
76 % |
|
cantbeat.ai: AI tools with deterministic outcomes
cantbeat.ai is an AI consultancy that builds tools with deterministic outcomes: constrained, testable systems that return the same answer for the same input,...
|
76 % |
|
Aver | Continuous validation for clinical bioinformatics pipelines
Aver benchmarks your clinical pipeline against public reference materials, detects when a change moves the numbers, and produces signed evidence your inspect...
|
76 % |
|
tracegrab — see what your API actually did
MCP-native runtime debugger. 31 tools let an AI agent set breakpoints, drive a scenario, and read a diagnosis — exceptions, N+1 queries, silent state loss, m...
|
76 % |
|
CleverQA — Every Failure Becomes a Diagnosed Defect · CleverQA
Monitors, crashes, server saturation, load tests and releases all feed one place: a defect with a likely cause attached, assigned and tracked in Jira, Linear...
|
75 % |
|
QAClan — browser and API regression, run on your machine
QAClan captures your browser flow and the API calls it fires, then replays both as one suite — on your machine, the same way every time.
|
75 % |
|
Change-Aware CI Analysis & Test Selection | DiffCI
DiffCI analyzes code changes, finds affected tests, and measures potential CI savings with conservative full-run fallbacks. Try the CLI or GitHub Action.
|
75 % |
|
Tapedeck · deterministic backtesting for TypeScript
An event-driven backtesting and paper-trading engine for TypeScript. No lookahead, fixed-point money, and it reports what it could not know. Run one free in...
|
75 % |
|
delta.engineer
The distance between a working prototype and production is where most AI-infra projects fail. This log measures that distance, daily — every number claim-lab...
|
75 % |
|
tracelint — structural linting for agent traces
tracelint is a deterministic, judge-free structural linter for tool-calling agent traces. Point it at the traces you already collect and fail CI on structura...
|
75 % |
|
Tracekit - Find What Broke in Production
Tracekit connects analytics, traces, errors, sessions, LLM calls, alerts, and dynamic logs so you can understand users and production issues.
|
75 % |
|
Veval — LLM Regression Testing
Veval captures real production traces and turns them into a regression baseline. When your prompt changes, know what broke before your users do.
|
75 % |
Detected social media networks