Safety Grades

Independent assessments of maintainer, permissions, supply chain, transparency, and incident history.

Grade	Item	Type	Score
A	pipecat	Agent	83
A	livekit-agents	Agent	83
A	microsoft-agent-framework	Agent	83
A	rasa	Agent	82
A	langgraph	Agent	82
A	junie	Agent	82
A	amazon-q-developer	Agent	82
A	cody	Agent	82
A	dust	Agent	82
A	github-copilot-workspace	Agent	82
A	sourcegraph-cody	Agent	82
A	tabnine	Agent	82
B	codium-pr-agent	Agent	74
B	llamaindex	Agent	73
B	n8n-ai	Agent	73
B	undermind	Agent	72
B	inkeep	Agent	72
B	sana	Agent	72
B	jasper	Agent	72
B	writer	Agent	72
B	harvey	Agent	72
B	voiceflow	Agent	72
B	cresta	Agent	72
B	nabla	Agent	72
B	finchat-fiscal	Agent	72
B	replit	Agent	72
B	le-chat-mistral	Agent	72
B	qwen-chat	Agent	72
B	dify	Agent	72
B	flowise	Agent	72
B	langflow	Agent	72
B	smolagents	Agent	72
B	mastra	Agent	72
B	letta	Agent	72
B	crewai	Agent	72
B	goose	Agent	72
B	warp	Agent	72
B	jules	Agent	72
B	autogen	Agent	72
B	refact	Agent	72
B	continue	Agent	72
B	cline	Agent	72
B	dia	Agent	72
B	chatgpt-atlas	Agent	72
B	exa	Agent	72
B	stagehand	Agent	72
B	firecrawl	Agent	72
B	tavily	Agent	72
B	clay	Agent	72
B	reka	Agent	72
B	gemini-deep-research	Agent	72
B	mistral-le-chat-agents	Agent	72
B	crew-studio	Agent	72
B	hex-magic	Agent	72
B	browserbase-agent	Agent	72
B	scite	Agent	72
B	elicit	Agent	72
B	builder-io-figma-to-code	Agent	72
B	retool-ai	Agent	72
B	sweep	Agent	72
B	fin-intercom	Agent	72
B	aider	Agent	72
B	openevidence	Agent	72
B	grok-deepsearch	Agent	72
B	zapier-central	Agent	72
B	skyvern	Agent	72
B	consensus	Agent	72
B	perplexity-spaces	Agent	72
B	v0	Agent	72
B	replit-agent	Agent	72
B	claude-code-agent	Agent	72
B	openai-operator	Agent	72
B	bolt	Agent	72
B	base44	Agent	71
B	otter	Agent	71
B	heygen	Agent	71
B	agno	Agent	71
B	openhands	Agent	71
B	plandex	Agent	71
B	browser-use	Agent	71
B	browserbase	Agent	71
B	ada-cx	Agent	68
B	you-com	Agent	68
B	zai-zhipu	Agent	68
B	potpie	Agent	68
B	cursor	Agent	68
B	glean	Agent	68
B	pi-ai	Agent	68
B	vapi	Agent	68
C	chatprd	Agent	58
C	outset-ai	Agent	58
C	kore-ai	Agent	58
C	ema	Agent	58
C	aisera	Agent	58
C	moveworks	Agent	58
C	paradox	Agent	58
C	aomni	Agent	58
C	rox	Agent	58
C	relevance-ai	Agent	58
C	regie-ai	Agent	58
C	gigaml	Agent	58
C	air-ai	Agent	58
C	hebbia	Agent	58
C	norm-ai	Agent	58
C	evenup	Agent	58
C	legora	Agent	58
C	paxton-ai	Agent	58
C	spellbook	Agent	58
C	openmic-ai	Agent	58
C	thoughtly	Agent	58
C	synthflow	Agent	58
C	polyai	Agent	58
C	parloa	Agent	58
C	cognigy	Agent	58
C	maven-agi	Agent	58
C	decagon	Agent	58
C	sierra	Agent	58
C	glass-health	Agent	58
C	abridge	Agent	58
C	daloopa	Agent	58
C	meetgeek	Agent	58
C	fellow	Agent	58
C	fireflies	Agent	58
C	pi-ai-inflection	Agent	58
C	cohere-north	Agent	58
C	doubao	Agent	58
C	minimax-agent	Agent	58
C	chatgpt-atlas-agent	Agent	58
C	agent-ai-marketplace	Agent	58
C	babyagi	Agent	58
C	autogpt	Agent	58
C	superagi	Agent	58
C	vellum	Agent	58
C	trae	Agent	58
C	imbue	Agent	58
C	supermaven	Agent	58
C	qodo	Agent	58
C	windsurf	Agent	58
C	comet	Agent	58
C	hyperbrowser	Agent	58
C	olvy	Agent	58
C	11x	Agent	58
C	rogo-ai	Agent	58
C	eudia	Agent	58
C	soraio	Agent	58
C	cosine	Agent	58
C	genspark	Agent	58
C	retell	Agent	58
C	julius	Agent	58
C	wordware	Agent	58
C	glif	Agent	58
C	multion	Agent	58
C	hyperwrite	Agent	58
C	cognosys	Agent	58
C	lindy	Agent	58
C	augment-code	Agent	58
C	bland	Agent	58
C	gumloop	Agent	58
C	agent-ai	Agent	58
C	lovable	Agent	58
C	devin	Agent	58
C	manus	Agent	58
C	artisan	Agent	54
C	rewind-ai	Agent	54
C	codegen	Agent	54
C	magic-dev	Agent	54
C	poolside	Agent	54
C	factory	Agent	54
C	tektonic-ai	Agent	54
C	beam	Agent	54
C	softgen	Agent	54
C	zencoder	Agent	52
C	anchor-browser	Agent	52
D	heyboss	Agent	42
D	youware	Agent	42

What the grades mean

A+ / A: Official vendor or well-maintained, scoped permissions, clean record. Safe defaults.
B: Sensible community pick. Review the install and permissions before granting sensitive access.
C: Usable but comes with caveats. Read the red flags.
D: Meaningful concerns. Only use in sandboxed or low-stakes contexts.
F: Avoid unless you know exactly what you're doing and have mitigations.

Grades are advisory. They update as the ecosystem changes. They do not replace your own review of the code and permissions before granting an agent access to sensitive systems.