The system
How the system worksThe DirectoryDispatchThe LedgerAI modelsAI agentsYour installationThe appThe research libraryWhat changes for your businessUse it from your softwareEverything it can do
Messaging
iMessageWhatsApp
Ads and leads
AdsCreativeLeads
Devices
Devices
Industries
RealtorsRestaurantsHome services
Explore
How it worksPermissions and limitsCompareSet upQuestionsText me

Is it LangChain? No. Here's the map.

Ask what this build is and the reflex answer is a name-check: "so it's LangChain?" It isn't, and working out why it isn't turns out to be the fastest way to explain what it actually is. The build is one deployed endpoint — POST

Is it LangChain? No. Here's the map.

Ask what this build is and the reflex answer is a name-check: "so it's LangChain?" It isn't, and working out why it isn't turns out to be the fastest way to explain what it actually is. The build is one deployed endpoint — POST /api/dispatch on a Cloudflare Worker — behind which sits a directory of 887 callable rows: functions, agents, HTTP recipes, whole external MCP servers absorbed as rows. Every call lands in a ledger. Authority is delegated by minting a scoped URL with a time-to-live, a use count, and a stated purpose. The public face is 2,156 articles with hash-chained claims and sources. None of that is a framework. All of it is what frameworks are sold as helping you build.

The frameworks are libraries. This is a running service.

LangChain, LangGraph, CrewAI, AutoGen — the names people reach for — are orchestration libraries you import into a Python process you host yourself. LangGraph "uses a graph-based approach to define and execute agent workflows" with nodes and edges. CrewAI assigns agents "distinct roles and goals." AutoGen has agents talk to each other through "asynchronous messages." They are ways of arranging LLM calls inside your own code.

The build has no graph definition anywhere. Routing is a directory lookup: a natural-language message or a {key, body} pair hits the one ingress, matches a row, and the row says what runs — a model, a function, an external API. The orchestration layer the frameworks sell is replaced by a table. That trade costs expressiveness (no compile-time cycles, no typed state machines) and buys something the frameworks don't have: the whole surface is data. A new capability is an inserted row, not a redeploy.

The protocol layer got standardized while nobody was watching

The honest comparison isn't to any framework. It's to the two protocols the industry converged on. MCP — open-sourced November 2024, donated to the Linux Foundation in December 2025, roughly 97 million monthly SDK downloads across Python and TypeScript — standardizes how an agent reaches a tool. A2A, which passed 150 supporting organizations in April 2026, standardizes how agents reach each other.

The build speaks MCP in both directions: /api/mcp exposes the directory to any MCP client, and MCP_IMPORT absorbs external MCP servers into directory rows. So the right sentence is: the build is not an alternative to MCP, it is a node that predates the registry layer and now plugs into it.

The closest commercial shape just got a death date

The nearest thing a lab ever shipped to this architecture was OpenAI's AgentKit: a hosted Agent Builder canvas plus a Connector Registry — "a unified administrative panel" that "consolidates data source access across ChatGPT and the OpenAI API," with credential management and audit logging. Registry of connectors, central credential store, audit trail. Same organs, different animal.

On June 3, 2026, OpenAI announced Agent Builder and the Evals platform leave the platform on November 30, 2026. ChatKit and the Connector Registry stay. The hosted visual builder — the piece positioned as the future of agent-building eight months earlier — lasted just over a year.

That is the structural argument for owning the kernel. A one-person build on a $5 Worker has no deprecation committee. The dependency it carries is narrower: model APIs and Cloudflare primitives, both of which have many substitutes behind one dispatch seam.

What the comparison actually returns

Run the layers separately and the answer stops being a name:

  • Orchestration: the frameworks do it in code; the build does it as directory data. Different bet, not a missing feature.
  • Tool connectivity: MCP won; the build imports and exports it.
  • Agent-to-agent: A2A is the emerging standard; the build does this today by treating other agents as rows.
  • Hosted builder: AgentKit was the analog; it is being sunset while its Connector Registry — the part that looks most like this build — survives.

The pattern across all four layers: the industry keeps keeping the registry and the log, and keeps churning everything else. This build started as the registry and the log. That's the answer to "is it LangChain": no — LangChain is scaffolding for building the organs this thing already runs.

References (5)
  1. A Detailed Comparison of Top 6 AI Agent Frameworks in 2026“LangGraph uses a graph-based approach to define and execute agent workflows”
  2. Everything your team needs to know about MCP in 2026 — WorkOS“roughly 97 million monthly downloads”
  3. modelcontextprotocol/registry (GitHub)“A community driven registry service for Model Context Protocol (MCP) servers.”
  4. OpenAI AgentKit: Build AI Agents Step-by-Step“consolidates data source access across ChatGPT and the OpenAI API”
  5. OpenAI AgentKit Deprecation Guide (2026)“ChatKit remains available.”