agent-reliability 开源项目推荐（314 个）

FailproofAI /failproofai

为 AI agent harness 提供可观测性与强制执行。借助策略执行捕获每一次运行，并保障运行时可靠性。

qualixar /superlocalmemory

世界上第一个在LoCoMo上实现74%检索和60%零LLM的本地AI内存。没有云，没有API，也没有数据流出你的机器。此外，C模式（LLM/云端）的LoCoMo比例为87.7%。有研究支持。arXiv： 2603.14588

yzhao062 /awesome-auditable-ai

Auditing AI agents: a curated list of papers, tools, datasets, benchmarks, and standards covering reliability, monitoring, failure attribution, and decision records.

trustabl /agent-reliability-analyzer

Agent Reliability Analyzer: deterministic scanner that finds and fixes AI agent reliability failures

dunetrace /dunetrace

The runtime reliability layer for AI agents.

trustabl /agent-reliability-rules

Deterministic agent reliability detection rules for the Trustabl Analyzer

ChaoYue0307 /awesome-loop-engineering

🔁 Build reliable recurring AI-agent systems: 1022 resources, 22 operational patterns, 22 loop contracts, 8 runtime starters, an interactive atlas, and a structured dataset.

trustabl /trustabl-aws

Agent Reliability Analyzer for AWS CodePipeline and CodeCatalyst: find agent reliability issues

trustabl /trustabl-action

Scan agents and auto-remediate agent reliability issues in GitHub Actions

zaxardery8011-design /soplint

Lint rules for your AI agent's discipline, not its code

trustabl /agent-dashboard

Agent dashboard runs reliability scans, skills, and network access probes.

qualixar /qualixar-os

Qualixar OS: The Universal OS for AI Agents. Claw-compatible. 12 topologies, Forge AI team designer, 24-tab dashboard, skill marketplace. PAPER: https://arxiv.org/abs/2604.06392

okgrit /done-is-a-claim

Your coding agent says done. Show the receipt. Evidence-first workflows, practical skills, and plugins for Claude Code and Codex.

kalibr-ai /kalibr-sdk-python

Stop overpaying to run your agents. Kalibr routes every request to lower-cost model and tool paths without degrading performance.

B0yko /agent-claimcheck

Catch AI agents that say "done" when the task isn't. Checks agent-trace/v1 traces with claim-evidence rules, LLM judges and a trained classifier, returns verified / false_success / unverifiable with calibrated confidence, and ships a 300-trace labelled benchmark.

maida-ai /maida

您的代理仍然返回正确的答案--但现在它调用了3倍的工具。Maida是人工智能代理的合并前行为回归门：基线代理跟踪、CI中的PR门、在合并前阻止行为回归。本地优先，没有云。

yzhao062 /auditable

Audit any agent decision across its past, present, and future, on one typed graph.

IgorGanapolsky /ThumbGate

ThumbGate Pre-Action Checks self-improve from ranked lessons and repeated failures, hard-block detected secret leaks, and block matches in strict mode.

ProvablyAI /sourcerykit

Python SDK for accurate and verifiable agent tool use. Agents verify that answers came from the right source and were not changed. Downstream agents detect 100% of errors and retry to achieve a 50% jump in answer accuracy.

tessaryai /tessary

Open-source agent reliability platform for discovering, investigating, and fixing production agent failures