AGNTCY

Executive Summary

AGNTCY (pronounced "agency") is the Linux Foundation's bet that the "Internet of Agents" — its own framing — needs more than a wire protocol — it needs the equivalent of DNS, PKI, and network monitoring. ^[1] The project ships four infrastructure layers for multi-agent systems: a decentralized Agent Directory for announce-and-discover, an Identity system with cryptographic verification, observability and evaluation SDKs, and SLIM (Secure Low-Latency Interactive Messaging) for many-to-many agent communication. All of it is designed to interoperate with A2A and MCP rather than replace them — the schema framework explicitly describes A2A agents and MCP servers. ^{[2] [3]}

Cisco's Outshift incubator open-sourced AGNTCY in March 2025 alongside LangChain and Galileo, then donated it to the Linux Foundation on July 29, 2025, with Cisco, Dell Technologies, Google Cloud, Oracle, and Red Hat as formative members and more than 65 supporting companies. ^{[4] [3]} Development is genuinely active — 44 of the org's 48 repositories saw pushes in 2026, and Directory shipped v1.4.0 in May 2026 — but adoption signal is weak: roughly 2,000 GitHub stars across the entire organization as of June 2026, and no AGNTCY story has ever cracked Hacker News's front page. ^{[5] [6]}

---

Product Overview

AGNTCY is consumed as a set of components, not a single install: you describe agents with OASF records, publish them to a Directory node, issue them verifiable identities, wire their traffic through SLIM, and instrument them with the observability SDK. A reference application, CoffeeAGNTCY, demonstrates the components end to end. ^{[2] [5]}

Key Capabilities

Product Surfaces

---

Technical Architecture

The components are polyglot and independently deployable: SLIM is written in Rust, the Directory and Identity in Go, OASF in Elixir, and the observability SDK in Python. ^[5] SLIM runs over gRPC and targets interaction patterns plain HTTP request/response handles poorly — group communication, streaming, and many-to-many messaging between agents. ^[2] The Directory is explicitly decentralized: any organization can run nodes and announce agents described by OASF records, the "Internet of Agents inventory" framing. ^[2]

A telling architectural decision: AGNTCY originally shipped its own agent protocol — the Agent Connect Protocol (ACP, not to be confused with IBM's identically abbreviated protocol) — and has since archived it, along with its workflow server and I/O mapper repos. The project now positions itself as infrastructure beneath A2A and MCP rather than a competing protocol. ^[5]

Key Technical Details

---

Strengths

• Heavyweight, vendor-neutral governance — Linux Foundation stewardship with Cisco, Dell, Google Cloud, Oracle, and Red Hat as formative members and 65+ supporting companies removes single-vendor risk on paper. ^[3]
• Fills a real gap the protocols don't — A2A defines how two agents talk; AGNTCY tackles discovery, identity, and observability at network scale, the layers any production multi-agent deployment eventually needs. ^[2]
• Interoperates instead of fragmenting — OASF describes A2A agents and MCP servers, and the project killed its own competing protocol rather than splitting the ecosystem. ^{[2] [5]}
• Genuinely active engineering — 44 of 48 repos pushed in 2026, with SLIM and Directory commits landing the day of this research (June 11, 2026) and a steady release train across components. ^[5]
• Identity is a differentiated bet — cryptographically verifiable agent identity issuable by any organization addresses a trust problem none of the messaging protocols solve. ^[3]

---

Cautions

• Adoption signal is thin — ~2,066 stars across the entire 48-repo org (top repo: 316) after 15 months, against 24K+ for the single A2A repo; corporate logos have not translated into visible grassroots usage. ^[5]
• Sprawl risk — six component categories, five implementation languages, and 48 repos is a lot of surface area for a project this young; HN's sole substantive reaction called it a "crazy sprawling framework." ^{[5] [6]}
• SLIM departs from HTTP — a bespoke gRPC-based messaging layer is an adoption hurdle when the rest of the agent ecosystem settled on HTTP-native protocols. ^[2]
• Strategy already pivoted once — the archived Agent Connect Protocol, workflow server, and I/O mapper show the original product thesis didn't survive contact with A2A; the surviving components could be re-scoped again. ^[5]
• Acronym hazard — AGNTCY's archived "ACP" (Agent Connect Protocol) is unrelated to IBM's ACP (Agent Communication Protocol, itself merged into A2A) and Zed's Agent Client Protocol, making 2025-era comparison content actively confusing. ^[5]

---

What Developers Say

Community discussion is strikingly sparse for a project with this many corporate backers: no AGNTCY submission has exceeded 3 points on Hacker News as of June 2026, and the mentions that exist are mostly side comments in other projects' threads. ^[6] That silence — fifteen months after launch — is itself a data point.

"Crazy sprawling framework. A directory of ACA and MCP tools. A bespoke low latency messaging system. Workflow schedulers." — jauntywundrkind, Hacker News ^[6]

"Great, another Agent 2 Agent protocol. One more of those and we'll have a complete set!" — mindcrime, in the A2A launch thread ^[7]

"A quick scan of the 'partners' for A2A includes many of the same groups that helped launch AGNTCY. Either they jumped ship or they're teaming up with everyone." — soccernee, Hacker News ^[7]

"Agntcy is a little more elaborate, but also notable as a massive AI tool that includes tool management. Donated to Linux Foundation from Cisco & many others." — jauntywundrkind, Hacker News ^[8]

---

Pricing & Licensing

Licensing model: Apache 2.0 across the org under Linux Foundation neutral governance; no commercial edition exists. ^{[3] [5]}

Hidden costs: Everything is self-hosted — running Directory nodes, SLIM message brokers, and identity services is your operational burden, and the five-language stack raises the maintenance bar. ^[2]

---

Competitive Positioning

Direct Competitors

When to Choose AGNTCY Over Alternatives

• Choose AGNTCY when: you operate many agents across organizational boundaries and need discovery, verifiable identity, and observability as shared infrastructure — and you have the platform engineering capacity to self-host it.
• Choose Google A2A alone when: you need agents to interoperate today; its Agent Cards cover point-to-point discovery without standing up directory infrastructure.
• Choose plain MCP + a registry when: your problem is agent-to-tool wiring inside one organization, not an inter-organization agent network.

---

Ideal Customer Profile

Best fit:

• Large enterprises (especially Cisco/Dell/Oracle/Red Hat shops) standardizing multi-vendor agent deployments that need discovery, identity, and audit infrastructure
• Platform teams building internal "agent networks" who want LF-governed, Apache 2.0 components instead of a proprietary control plane
• Vendors that want their agents discoverable and verifiable across organizational boundaries

Poor fit:

• Startups shipping a single agent product — A2A plus MCP covers them with far less operational surface
• Teams wanting a managed service; every AGNTCY component is self-hosted
• Anyone allergic to early-stage churn — Identity is still v0.0.x and the project has already archived one product line ^[5]

---

Viability Assessment

The corporate-backing-to-community-traction gap is the defining tension: 65+ supporting companies and five formative members would normally signal a winner, but fifteen months in, the grassroots adoption that made MCP and A2A defaults has not materialized. ^{[3] [5]} The graceful retreat from the protocol race shows the project can self-correct, which counts for something at this stage. ^[5]

---

Bottom Line

AGNTCY is the most institutionally credible attempt at the layer above the agent protocols — discovery, identity, and observability for networks of A2A and MCP agents, under Linux Foundation governance with real engineering velocity behind it. But infrastructure standards win on usage, not membership rosters, and AGNTCY's usage signal is faint: ~2K org-wide stars, near-zero community discussion, and one already-abandoned product line. ^{[5] [6]}

Recommended for: Enterprise platform teams building multi-vendor, cross-organization agent networks who want open, vendor-neutral discovery and identity infrastructure and can self-host it.

Not recommended for: Single-product agent teams (use A2A + MCP directly), anyone needing a managed service, or teams that require proof of production adoption before betting on infrastructure.

Outlook: Watch whether the A2A project blesses AGNTCY's Directory and Identity as its official discovery/identity story — that would make AGNTCY the default plumbing of the agent ecosystem. Absent that, it risks becoming a well-governed stack waiting for a network that standardized elsewhere.

---

Research by Ry Walker Research • methodology