frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Agent Trust – Cryptographic identity and reputation for AI agents

https://github.com/kanoniv/agent-trust
1•dreynow•1h ago

Comments

dreynow•1h ago
I've been building multi-agent systems and kept running into the same problem: every framework trusts agents by default. There's no identity verification, no way to learn which agent performs well on which task, and no mechanism to revoke permissions when something underperforms. You only find out something went wrong after reading traces.

I built Agent Trust to explore a different approach: treat agents like services with identity, permissions, and track records. It's a Python SDK (MIT, 135 tests) that adds several layers:

Identity – Each agent gets an Ed25519 keypair and a DID (did:agent:). Every action is cryptographically signed.

Delegation – Scoped permissions with caveats and expiry. An agent can only do what it's been explicitly allowed to do.

Reputation – Computed from verified outcomes, not self-reported metrics.

Routing – UCB-based selection picks agents based on past performance, balancing exploration and exploitation.

Enforcement – Permissions can be restricted or revoked at runtime. Cryptographic, not advisory.

pip install kanoniv-trust

from agent_trust import TrustAgent

trust = TrustAgent() # SQLite, zero setup

trust.register("researcher", capabilities=["search", "analyze"]) trust.register("writer", capabilities=["draft", "edit"])

trust.delegate("researcher", scopes=["search"], expires_in=3600) trust.delegate("writer", scopes=["draft", "edit"])

trust.observe("researcher", action="search", result="success", reward=0.9)

trust.authorized("researcher", "search") # True trust.authorized("researcher", "analyze") # False — not delegated

best = trust.select(["researcher", "writer"], action="search")

The part I've found most useful: agents can read their own verified track record before acting.

ctx = trust.recall("researcher") print(ctx.guidance) "researcher excels at search (95% success). Weaknesses: none observed. Recommendation: High confidence for search tasks."

This injects a summary of past outcomes — success rate, strengths, weaknesses — directly into the prompt. Agents adapt based on actual performance without retraining. It's a simple form of in-context reinforcement learning, and it's the thing that surprised me most: agents genuinely behave differently when they can see their own track record.

There's also a dashboard (Observatory) for visualizing reputation scores and delegation graphs, and integrations for LangChain and CrewAI.

I looked at Langfuse, AgentOps, and similar tools — they're good at tracing but stop at observation. This tries to close the loop: identity, verified history, and decision-making in one system.

It's early but usable today. I'd especially appreciate feedback from people running multi-agent systems in production.

GitHub: https://github.com/kanoniv/agent-trust

When is it ok to slop your colleagues?

https://nickheiner.substack.com/p/when-is-it-ok-to-slop-your-colleagues
1•NTH•39s ago•0 comments

Sam Altman is under fire from critics again for 'disgusting' AI remarks

https://www.indy100.com/science-tech/sam-altman-openai-chatgpt-latest
1•dmitrygr•46s ago•0 comments

Show HN: CLI tool for generating AI images

https://github.com/michaeldmueller/picasso
1•lurkingllama•1m ago•0 comments

Internet and Airstrikes: Tracking Iran's Extended Communication Blackout

https://www.kentik.com/blog/internet-and-airstrikes-tracking-irans-extended-communication-blackout/
1•oavioklein•2m ago•0 comments

Show HN: Crew Chief – OBD2 diagnostics via cheap Bluetooth scanner

https://crewchief.cc/
1•meandave•2m ago•0 comments

How do you handle repetitive developer support questions?

1•crawldesk•2m ago•0 comments

CUDb, a lightweight GPU-native DB Engine

https://github.com/andre-git/cudb
1•andre-hn•4m ago•1 comments

Infosec Survival Guide Orange Book

https://www.blackhillsinfosec.com/prompt-zine/prompt-issue-infosec-survival-guide-orange-book/
1•QuantumAtom•5m ago•0 comments

MCP servers mass-forked and republished – supply-chain attack vector

1•ultrafox42•6m ago•0 comments

Ask HN: Human psychology of non-AI-native users

1•ajaystream•6m ago•1 comments

A live adversarial benchmark crowdsourced from domain experts

https://www.rusmarterthananllm.com/
1•camillemolas•7m ago•1 comments

Show HN: I built an audiobook player that syncs with your physical book

https://earleaf.app/blog/introducing-earleaf
1•arcadianalpaca•8m ago•0 comments

Accelerated north–east shift of the global green wave trajectory

https://www.pnas.org/doi/10.1073/pnas.2515835123
1•PaulHoule•8m ago•0 comments

The OWASP MCP Top: A Security Framework for AI Agent Tool Integration

https://mcpblog.dev/blog/2026-03-15-owasp-mcp-top-10
1•algis-hn•8m ago•0 comments

Keep Hyper-Scale Datacenters Out of the Desert (March 2026)

https://alec.is/posts/keep-hyperscale-datacenters-out-of-the-desert/
1•arm32•10m ago•0 comments

Show HN: AI agent deploys an edge AI model on a microcontroller via MCP

https://es617.github.io/2026/03/16/edge-ai-mcp.html
1•es617•11m ago•0 comments

Closing Arguments Begin in Twitter Trial Accusing Musk of Driving Down Stock

https://www.law.com/therecorder/2026/03/17/closing-arguments-begin-in-twitter-shareholder-trial-a...
1•1vuio0pswjnm7•11m ago•0 comments

The Context Must Flow – Permit MCP Gateway Explainer [video]

https://www.youtube.com/watch?v=pLQCG31HSK8
1•CruddyDoctor229•11m ago•0 comments

The Ugliest Airplane: An Appreciation

https://www.smithsonianmag.com/air-space-magazine/ugliest-airplane-appreciation-180978708/
1•randycupertino•12m ago•0 comments

Node.js worker threads are problematic, but they work great for us

https://www.inngest.com/blog/node-worker-threads
1•goodoldneon•13m ago•0 comments

Show HN: BulkHead – iOS File Manager for SFTP, SMB, WebDAV, and S3

https://www.oddinks.com/bulkhead/
1•xydac•13m ago•0 comments

Tokens Are Not Securities

https://www.bloomberg.com/opinion/newsletters/2026-03-18/tokens-are-not-securities
1•toomuchtodo•14m ago•0 comments

Intel enables Precompiled Shader Delivery in new driver

https://videocardz.com/newz/intel-enables-precompiled-shader-selivery-on-arc-b-series-and-core-ul...
1•davikr•14m ago•0 comments

Comprehension Debt

https://addyosmani.com/blog/comprehension-debt/
1•ragall•15m ago•0 comments

Three More Jury Notes in Social Media Trial Focus on YouTube Witness, Father

https://www.law.com/2026/03/17/two-more-jury-notes-in-social-media-addiction-trial-focus-on-youtu...
2•1vuio0pswjnm7•15m ago•0 comments

The problem is not generating code. It is trusting the change

https://ferrify.uk/
1•gh7941•16m ago•2 comments

Show HN: Trickle – See PyTorch tensor shapes inline in VSCode as you code

https://github.com/yiheinchai/trickle
2•chaiyihein•17m ago•0 comments

Xiaomi MiMo-V2-Omni: See, Hear, Act in the Agentic Era

https://mimo.xiaomi.com/mimo-v2-omni
1•gainsurier•18m ago•0 comments

GitHub and Claude Are Down Three Out of Four Days

https://www.aakash.io/tech-chase/github-and-claude-are-down-three-out-of-four-days
1•ahstilde•18m ago•0 comments

Recovery of the first fiber-optic submarine cable across an ocean – TAT-8

https://www.submarinenetworks.com/en/nv/insights/recovery-of-the-world-s-first-fiber-optic-submar...
2•marklit•20m ago•0 comments