We Debug Our AI Harness on Weak Models on Purpose
July 20, 202616 min read
How we use weak, cheap models to find bugs strong models hide - and what a nightly 26-task benchmark says about which ones you can trust with real work.
Read more
Blog
Latest news, updates, and insights from the Archestra team
New: Apps or "we built ChatGPT Sites the right way"
July 10, 20263 min read
Archestra Apps are mini apps you build by chatting. They render in Archestra, in Claude, or by URL — and reach your data through MCP with per-user credentials, audit logs, and guardrails.
Read more
New: Migrating from Claude, OpenClaw, Hermes to centralized platform
July 9, 20263 min read
The migration kit inventories an existing agentic setup of personal agents and migrates it to a centralized agentic runtime in Archestra for further collaboration in your org.
Read more
New: Collaborative Projects and Persistent Files
July 8, 20262 min read
Archestra agents now have a collaborative workspace: files that persist across chats and runs, and Projects that keep chats, files, scheduled tasks, and instructions together.
Read more
New: Connect Coding Agents to LLM & MCP Gateways in 45 seconds
July 7, 20262 min read
Archestra 1.3 lets every engineer wire their coding agents into LLM & MCP Gateways in one step — governance, cost management, logs, access rules, and shared skills.
Read more
New: Skills and Code Execution
July 6, 20263 min read
Archestra 1.3 "Lyra" brings Skills to the platform: a central catalog, sandboxed code execution, and governance.
Read more
Archestra now supports Okta's Cross-App Access
June 23, 20264 min read
Archestra is joining Okta's Cross-App Access (XAA) ecosystem, bringing identity-based, per-user access to AI agents and MCP without shared API keys or repeated consent flows.
Read more
A Drizzle Migration Linter for Zero-Downtime Postgres Deploys
June 5, 20266 min read
Why zero-downtime deploys need expand/contract migrations, what Django got right, and how we built a Drizzle migration linter for CI.
Read more
Archestra.AI Announces $10M Seed 🎉
June 2, 20263 min read
Ten months after our pre-seed, we're announcing a $10M Seed led by 20VC — and what we learned getting here.
Read more
Archestra Now Supports All Types of LLM Auth
May 5, 20266 min read
How Archestra authenticates LLM gateway calls across provider keys, virtual keys, OAuth, and enterprise identity providers.
Read more
Let's talk about AI slop
April 17, 20265 min read
Is it the end of open source we know and love?
Read more
Enterprise-Managed Authorization for MCP
March 30, 202613 min read
Centralized MCP server access control through enterprise identity providers.
Read more
Building Enterprise-Ready MCP Servers with JWKS and Identity Providers
February 11, 202610 min read
How to integrate MCP servers with enterprise identity providers using JWKS for scalable, stateless token validation.
Read more
A Developer's Guide to MCP Authentication
February 11, 20269 min read
The standards, protocols, and patterns behind secure AI tool access — OAuth 2.1, PKCE, CIMD, and more.
Read more
How to Run OpenClaw Securely
February 4, 202610 min read
OpenClaw is a security nightmare out of the box. Here is how to lock it down with Archestra.
Read more
A self-hosted Movie Recommendation Agent with LightRAG and MCP
January 12, 20267 min read
How I built a personal movie expert with LightRAG and Archestra that learns from my feedback
Read more
Brew by Weight? Brew by AI!
December 28, 202512 min read
A Christmas project of making AI brew an excellent coffee.
Read more
One Month: From v0.0.1 to v0.6.25
December 9, 20253 min read
379 PRs, 2 new team members, and 4 major features - here's what we've been building at Archestra
Read more
Archestra Platform: The Open-Source Agentic Gateway for Secure AI
October 13, 20253 min read
The first release of the Archestra Platform, our open-source agentic gateway built to secure your AI agents
Read more
The Archestra Dual LLM Pattern: Teaching AI Agents to Play a Guess Who? Game with Untrusted Data
October 13, 20256 min read
How we developed a new security pattern inspired by a childhood game to protect AI agents from prompt injection attacks
Read more
What is a Prompt Injection?
October 13, 20259 min read
A deep dive into prompt injection attacks, why even the smartest AI models are vulnerable, and how Archestra Platform protects your AI agents from this critical security threat.
Read more
Dominik, welcome post
September 18, 20253 min read
Dominik Broj joining Archestra as a Founding Engineer bringing 10+ years of experience as Engineer at such companies as Grafana Labs, DataArt and Sportradar.
Read more
Thank you, 100 MCP Server builders!
September 9, 20253 min read
Archestra Catalog is supported by 100+ MCP servers, thank you!
Read more
Joining the CNCF and The Linux Foundation
August 20, 20252 min read
Archestra has joined the Cloud Native Computing Foundation (CNCF) and The Linux Foundation to reinforces our commitment to open-source.
Read more
Announcing $3.3M Pre-Seed
August 14, 20253 min read
Archestra raises $3.3m pre-seed funding to build the security-first platform that enables enterprises to safely leverage AI agents and MCP
Read more
Why We Founded Archestra
August 11, 20253 min read
Why can't an AI draft 150 personalized emails to investors for my new startup?
Read more
Notes
A Catalog of Prompt Injection TechniquesThe A2A protocol in 5 bullet pointsHow to put a budget on an agentThe minimum agent observability stackTriggers, schedules, webhooks: how an agent decides to runMost "agents" are workflows with a triggerMost AI governance is a docApache APISIX AI Gateway + Archestra: Open Source on Top of Open SourceAudience-bound tokens, in one exampleAuditing the MCP supply chainWhy "fully autonomous" is rarely the answerWhen the LLM is not the expensive partAWS Bedrock + Archestra: Model Access vs. Agent PlatformClaude Desktop vs Archestra for MCPCloudflare AI Gateway + ArchestraWhy an AI security project belongs in the CNCFThe confused deputy problem, MCP editionThe context layer that pays for itselfData exfiltration via Markdown image tagsThe dual-LLM pattern in 200 wordsEmail is the most underrated agent triggerSo you're the 'MCP person' at your company nowA 10-point enterprise readiness checklist for any MCP serverEntra On-Behalf-Of flow for MCP, explainedHelicone + Archestra: LLM observability vs agent observabilityWhen to reach for a Claude hook vs a skillHow many MCP tools is too many?Human-in-the-loop for every write toolID token vs ID-JAG vs MCP access tokenWhat is JWKS, and why your MCP server needs it"Knowledge base" vs "RAG": what is the actual difference?Kong AI Gateway + Archestra: Same Building, Different FloorsHow to actually learn MCP in a weekThe Lethal Trifecta in one diagramLiteLLM Archestra: Do You Need Both? The Layer Each One Ownsllms.txt is the new robots.txt for MCP vendors and AI toolingMartian router Archestra: cost-aware model routing, identity-aware tool policyMastra Archestra: which layer does whatHow the Archestra MCP server quality score worksMCP Code Mode: collapsing tool round-tripsMCP Context Window Too Many Servers: Why 100 MCP Servers Crush Your AgentMCP for Non-Technical Users: Fixing the Install UXWhat an MCP gateway buys youMCP production issues: what actually breaks in real deploymentsMCP OAuth 2.1 Quick ReferenceRemote MCP server vs local MCPWhat does MCP 'sampling' actually do?Where Do MCP Server Secrets Actually Live?MCP Security Checklist: 7 Pre-Install Checks for Community MCP ServersMCP server config sync: stop the drift, centralize the gateway'MCP server' and 'MCP client' are confusing namesThe Best MCP Servers That Changed How Teams WorkA short history of the MCP spec versionsMCP STDIO Command Injection: The Bug Class Nobody Is PatchingEvery 10th MCP server is one personWhen MCP makes sense the second time aroundWhat a real MCP audit trail for tool calls looks likeMCP tool description drift: how to detect and prevent itMCP Tool Description: Why It Is a Prompt, Not DocumentationMCP Tool Naming Conventions for Reliable Agent Tool SelectionMCP tools, resources, prompts: the three primitives explainedMCP vs CLI toolsMCP vs function callingMCP vs LangChain toolsMCP vs REST API: the actual answerModel Router vs LLM ProxyThe five tasks every business automates firstWhy multi-agent demos die in productionn8n vs Archestra for agentic workflowsnpx MCP server install is not installationThe MCP OAuth 2.1 sequence problemOkta Entra MCP authorization: implementer comparisonOpenRouter Archestra: One Model API Meets Agent SecurityThe honest tradeoff for OSS agent stacksPin your MCP server versionsPKCE MCP: Why Every MCP Client Requires ItPortkey + Archestra: prompt-layer vs tool-layer guardrailsPractical MCP Use Cases Beyond Coding AgentsWhen to run a private MCP registryProgressive disclosure MCP: modeled vs measured savingsWhen the agent itself catches the injectionHow one production AI bot got owned in 10 secondsA field guide to prompt injection attacks (with examples)Prompt injection vs jailbreakCapacity is now the dominant agent failure modeHow to Rotate MCP Server Credentials Without a Manual NightmareThree reasons to sandbox every MCP serverSCIM for AI agentsClaude Skills vs MCP, when each one winsState of MCP, mid-2026 snapshotMCP stdio vs HTTP: When to Use Each TransportHow to test MCP servers: three layers that actually workWhen fast AI builds meet regulated industriesThe actual case for MCP, for skepticsThin MCP vs Fat MCP: Why Wrapping REST Endpoints FailsMost production failures are tool-definition failuresMCP Large Data Transfer: Use Resources and Pre-Signed URLsTrueFoundry Archestra: Serving the Model vs. Governing the AgentMCP prompts, resources, and sampling: the underused primitivesVercel AI Gateway + ArchestraWhat is MCP, in one paragraph?What is WebMCPWhat's Next After MCP? The Honest AnswerFine-tune vs Tool Call: When to Use Each for LLM AgentsAgent liability, the question nobody answersWhy agentic AI costs what it costsWhy agents feel solid at first, then quietly get worseAgent Tool Allowlist: Why Allowlists Beat BlocklistsOAuth for AI Agents: Why the Spec Wasn't Built for ThemWhy Official MCP Servers Are Thin: The Auth Problem Behind Half-Baked SaaS Wrappers