Never wanted my devices on 24/7 before AgentMesh changed that lol

Introducing AgentMesh (agentmesh-proxy), the governance proxy for every AI tool your team uses. 85% cache hits. 75% lower cost. For a team spending $48K/yr on AI tools, that's a $12K bill instead with zero code changes. Every AI tool your team uses talks to the LLM on its own and no shared cache, no shared budget, no audit trail. So I spent the last few months building the layer that was missing. I just published 85% of Our LLM Calls Never Reach the Model medium.com/p/85-of-our-llm-c…

This isn't a demo repo. It's shipped end-to-end: - PyPI - agentmesh-proxy - Docker - anilsprasad/agentmesh - Hugging Face Space (live) - Chrome Web Store (in review) - 13/13 e2e tests passing - Apache 2.0, Ed25519-signed audit log Built solo.

The interesting part: most repeated prompts aren't exact duplicates. Engineers rephrase. Add "you are a senior architect." Switch to British spelling. Wrap it in markdown. AgentMesh strips all of that, then compares embeddings. Two differently-worded prompts → one cache hit.

Real numbers, no API keys, run it yourself: pip install agentmesh-proxy sentence-transformers python examples/benchmark.py 20 requests · 5 topic clusters · 4 phrasings each: 85% cache hit rate 75% lower cost 3 misses (cold start only)

AgentMesh is an OpenAI-compatible proxy. Point your tools at it: ANTHROPIC_BASE_URL=http://localhost:8080 OPENAI_BASE_URL=http://localhost:8080/v1 Every call now flows through: → 3-layer cache → per-team token quotas → cheapest-model routing → tamper-evident audit log

75% cost save, 85% cache hits. $48,000 token cost reduced to $12,000 Every AI tool your team uses talks to the LLM alone. No shared cache. No shared budget. No audit trail. So I built the missing layer. AgentMesh is one proxy in front of Claude Code, Copilot, ChatGPT, Gemini your own agents. Zero code changes. How it works 🧵

Real numbers — no API keys, run it yourself: pip install agentmesh-proxy sentence-transformers python examples/benchmark.py 20 requests · 5 topic clusters · 4 phrasings each: 85% cache hit rate 75% lower cost 3 misses (cold start only)

AgentMesh is an OpenAI-compatible proxy. Point your tools at it: ANTHROPIC_BASE_URL=http://localhost:8080 OPENAI_BASE_URL=http://localhost:8080/v1 Every call now flows through: → 3-layer cache → per-team token quotas → cheapest-model routing → tamper-evident audit log

Are your AI agents just passing notes, or are they actually collaborating? 🤖💬 Multi-agent coordination is evolving into two distinct architectural patterns: 1/ Task-Oriented 2/ Goal-Oriented (Simulation) Insight from Agent Mesh 📖 #databookclub #agentmesh #agenticAI #LLMs

Wow, just joined the AgentMesh community 🔥 This AI cross-chain intent project feels like catching lightning early

Why I'm staying bullish on AgentMesh 1/4 AI just needs my goal and handles everything 2/4 Community feels real, not pumped 3/4 Early entry window looks promising 4/4 Let's ride this wave together!

Use AgentMesh as a reputation service for agentic supply chain: agentmesh.knostic.ai/ VirusTotal scan referenced: virustotal.com/gui/file/1a55… Thank you for the partnership, Bernardo Quintero.

I built an interactive calculator for this — plug in your own context size, task frequency, and model. Try it yourself and see where your current setup lands. Built this experiment as part of my work on AgentMesh — multi-agent infra where context costs compound fast.

deployed AgentMesh ($MESH) contract address is 0x6E6E93dd9Eb21bb3FA207DA5D6069C4D86014bA3 view token: bankr.bot/launches/0x6E6E93d… concept: AgentMesh is a coordination and reputation layer for autonomous AI agents on Base. agents stake MESH to join task pools, collaborate onchain, and earn reputation-weighted rewards. the token captures value from agent coordination fees and protocol governance.

Anthropic「Zero Trust for AI Agents」の解説記事: 核となる主張は、AIエージェントを単なる補助ツールではなく「IDを持つ業務主体」として扱うべき、というもの。 フロンティアAIが脆弱性の発見から悪用までの時間を「数か月から数時間」に圧縮する時代において、従来のセキュリティ基準では不十分で、導入時の最低ライン（Foundation）自体を引き上げる必要がある、と。 セキュリティ基準は3段階に整理されている。 ・Foundation: 短命トークン、暗号学的ID、Deny by Default、エージェント隔離、包括的ログ ・Enterprise: mTLS、動的権限、サンドボックス、改ざん困難な監査証跡、異常検知 ・Advanced: ハードウェアバインド、継続的認可、JIT/JEA、confidential computing 実装アーキテクチャは4層構造で推奨。 ・Agent OS（目的・許可行動・禁止行動・ポリシー） ・AgentMesh（ID・通信経路・MCP/ツール/SaaSの接続） ・Agent Runtime（実行環境・権限・停止・隔離） ・Agent SRE（監査証跡・復旧・障害対応） 主要な攻撃面は6つ。Prompt Injection（外部データの隠れた命令化）、Tool Poisoning（MCPサーバー・スキーマの汚染）、Tool Chaining（正規ツールの危険な組み合わせ）、Identity Abuse（権限の過度な継承・委譲）、Memory Poisoning（メモリ内の悪意ある情報残存）、Supply Chain Risk（AI-BOM・モデル・依存関係の脆弱性）。 キーとなる転換点は「Least Privilege から Least Agency へ」。権限の広さだけでなく、自律実行の広さまで制限する。どのツールで、どの頻度で、どこまで自律実行できるかを設計する、という発想。 運用面ではAgentic SOARの設計原則も提示。証拠収集・ログ相関・一次トリアージ・影響範囲の仮説は自動化する一方、封じ込め判断・顧客通知・外部開示は人間に残す。dwell timeとcoverageを測定指標とする。 「AI時代のセキュリティは禁止ではなく統制であり、防御側もAIを使わなければ対抗不可能」というのが記事全体のメッセージ。

𝐭𝐨𝐝𝐚𝐲 𝐢 𝐡𝐢𝐭 500 𝐩𝐨𝐢𝐧𝐭𝐬 𝐨𝐧 𝐚𝐫𝐜 𝐡𝐨𝐮𝐬𝐞. 𝐨𝐧𝐞 𝐦𝐨𝐧𝐭𝐡, 𝐭𝐰𝐨 𝐛𝐮𝐢𝐥𝐝𝐬. It's exactly a month today, since i started contributing on @arc and one month in on arc ?? here's what happened🧵: _________ it started with a 20-day arc build sprint. no roadmap. just me learning and shipping. wrote content, deployed contracts, broke things, and learned the stack from scratch. that sprint taught me more about arc's primitives than anything else could have. then came the lablab.ai x circle agentic economy hackathon on arc. i built agentmesh, a multi-agent research economy where autonomous agents pay each other per task in real-time USDC on arc testnet. built the entire backend with circle developer controlled wallets, tried to integrate nanopayments, debugged a few thing but couldn't submit in time caux of my health, but shipped the product regardless. now i'm co building "settlo". an all-in-one autonomous fintech protocol on arc. deposit, pay, save, and borrow in USDC — from any chain, any token. today, i hit 500 points on arc house and here's what I've really learnt so far: >circle dev controlled wallets and how agents hold real balances >nanopayments — off-chain auth, batched onchain settlement, sub cent economics >x402 payment standard for agent to agent commerce >CCTP V2 for cross-chain USDC movement >arc testnet deployment and the block explorer >why arc's gas model changes everything for agentic finance this is just the beginning for me, and can't wait for settlo to go live soon! just buidl on arc 🫵

We released AgentMesh for free, a VirusTotal for extensions, skills, etc.: agentmesh.knostic.ai And our product to secure coding agents and their supply chain is free up to five users: knostic.ai

We opened up AgentMesh for free. It’s like a VirusTotal but got extensions, skills, etc. (agentmesh.knostic.ai). You can also check out how we secure coding agents and their supply chain, on the IDE, if you like. Free up to 5 licenses. knostic.ai

We opened up AgentMesh for free. It’s like a VirusTotal but got extensions, skills, etc. (agentmesh.knostic.ai). You can also check out how we secure coding agents and their supply chain, on the IDE, if you like. Free up to 5 licenses. knostic.ai