Free tools Windows power users keep installed
One-click scans. No signup required.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The standard, open framework for building AI agents from Cisco, LangChain and Galileo is called AGNTCY, but AGNTCY is not one agent-building framework or a finished universal standard. It is an open interoperability stack for describing, discovering, identifying, connecting and operating agents built with different frameworks and deployed across organizations.
AGNTCY is already public. The project launched in March 2025, initial components appeared during 2025, and Cisco says AGNTCY was donated to the Linux Foundation in July 2025. By August 2026, its public GitHub organization lists active repositories, documentation and a technical steering structure.
Key takeaways
- AGNTCY is an open-source interoperability project for multi-agent systems, not a replacement for LangChain, LangGraph, MCP or A2A.
- OASF describes agent capabilities and metadata, while the Agent Directory publishes and discovers agent records.
- ACP was proposed as a REST/OpenAPI interface for invoking remote agents, but the original ACP specification repository was archived on April 11, 2026.
- AGNTCY is most relevant when agents cross framework, team or organizational boundaries; it is usually unnecessary for a single-agent proof of concept.
- Discovery and identity can help establish what an agent is and where it came from, but neither proves that the agent is safe, accurate or competent.
What is the standard, open framework for building AI agents from Cisco, LangChain and Galileo?
The project is AGNTCY, pronounced like “agency,” and the most accurate description is an open ecosystem for agent interoperability. Cisco, LangChain and Galileo introduced AGNTCY as an “Internet of Agents” initiative: a collection of schemas, directories, communication components, identity tools and testing projects designed to help independently built agents work together.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →The original announcement came from Galileo on March 5, 2025, and described an open collective involving Cisco, LangChain, Galileo and other participants. Galileo’s launch announcement framed the problem as a need to discover, compose, deploy and operate agents across fragmented ecosystems.
Popular on Amazon
- PLUG-AND-PLAY GIGABIT MANAGED SWITCH: 8 x 1Gbps auto-negotiating ports work the moment you plug in — full-gigabit speed over Cat5e/Cat6 cabling.
- MANAGED, WITHOUT THE COMPLEXITY: Easy Smart web GUI on Windows, Mac or Linux — no app or Windows-only utility, unlike many competing switches.
- SEGMENT & PRIORITIZE TRAFFIC: Up to 64 VLANs, QoS, IGMP snooping and port mirroring keep voice, video and data fast, secure and organized.
- BUILT-IN PROTECTION: Auto DoS prevention, loop detection, broadcast storm control and cable test keep your network stable and easy to troubleshoot.
- RELIABLE 24/7 BACKBONE: Rugged fanless metal housing runs cool and silent at 0 dBA — the managed switch trusted in homes, offices and small business.
The wording “coming from Cisco, LangChain and Galileo” is now stale. AGNTCY launched publicly in March 2025, Cisco says the project was donated to the Linux Foundation in July 2025, and the current project organization presents AGNTCY as part of the Linux Foundation. Cisco announced its intent to acquire Galileo on April 9, 2026, then said the acquisition was completed on May 22, 2026. Galileo is therefore best described as an original founding contributor and now a Cisco-owned company, not simply an independent partner alongside Cisco.
That history matters, but it does not turn AGNTCY into a Cisco-only product. The project’s current governance and repository structure should be evaluated separately from the commercial relationship among its original participants.
Why is interoperability difficult for AI agents?
Interoperability is difficult because an agent is more than a model endpoint. An agent may include planning logic, tools, memory, state, permissions, streaming behavior, authentication, framework-specific messages and organization-specific policies.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Consider a customer-support agent that needs help from an invoice-review agent:
- The customer-support agent must discover an invoice agent that claims the right financial skills.
- The customer-support system must determine where the invoice agent is deployed and which protocol it supports.
- The two systems must establish identity and authorization before sensitive invoice data is exchanged.
- The caller must understand how to invoke the remote agent, receive progress, handle errors and interpret the result.
- Both teams need provenance, tracing, evaluation and audit records for the interaction.
Without shared conventions, each integration becomes a private adapter. Capability discovery may live in documentation or hard-coded configuration. Message formats may be tied to one framework. Authentication may work only inside one company. An agent can appear available while its tools, model, permissions or API have already changed.
AGNTCY targets this infrastructure gap. AGNTCY does not attempt to standardize every agent’s reasoning process. Instead, AGNTCY seeks common ways to describe agents, find them, identify them, connect them and test or operate them across environments.
What does AGNTCY include?
AGNTCY is a collection of components rather than a single runtime. The project’s core working-group charter and public repositories identify several pieces of the intended stack.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
| Component | Primary job | What it does not do by itself |
|---|---|---|
| OASF | Describes agent capabilities, skills, domains, deployment information and related metadata. | Run an agent or make an agent capable of completing a task. |
| Agent Directory | Publishes, stores and discovers agent records using structured metadata. | Guarantee that a discovered agent is safe, competent or available. |
| ACP | Proposed REST/OpenAPI interface for configuring and invoking remote agents. | Provide a universally adopted or permanently stable communication standard. |
| SLIM | Secure Low-Latency Interactive Messaging, one of AGNTCY’s communication components. | Replace application-level authorization, policy or behavioral evaluation. |
| Identity | Tools for onboarding and creating or verifying identities for agents, MCP servers and multi-agent systems. | Prove that an agent’s output is correct or that its behavior is safe. |
| CSIT | Continuous System Integration Testing for AGNTCY projects. | Guarantee compatibility with every external framework or deployment. |
| CoffeeAGNTCY | Reference application intended to demonstrate how project components work together. | Serve as proof of broad production adoption. |
The project also includes workflow and bridge-related work. The exact boundaries between components can evolve, so production teams should check the current repository and documentation rather than treating an early architecture diagram as a fixed specification.
How does OASF describe an agent?
OASF, the Open Agent Schema Framework, is a structured schema system for representing an agent’s capabilities, interactions and metadata. An OASF agent record can include skills, domains, identifiers, deployment details and extensible modules. The OASF repository is the authoritative place to check the current schema, extensions and licensing.
OASF is metadata, not an agent runtime. An OASF record does not give an agent a reasoning capability, connect a missing tool or validate the quality of a model. OASF gives another system a machine-readable description of what the agent claims to do.
A simplified illustrative record might look like this:
Also popular on Amazon
- 8 Gigabit Ethernet Ports: Expand your network with 8 high-speed ethernet ports for enhanced connectivity and performance
- Easy Smart Management: Manage and configure your network effortlessly via a web interface or free software
- Support VLAN: Segment traffic with up to 32 VLANs simultaneously out of 4K VLAN IDs for better security
- Network Monitoring: Monitor your network effectively with port mirroring, loop prevention, and cable diagnostics
- IGMP Snooping: Enhances multicast application performance for improved network efficiency
{
"name": "invoice-review-agent",
"description": "Reviews invoices for policy violations",
"skills": [
"financial document analysis",
"policy compliance"
],
"domains": [
"finance",
"accounts payable"
],
"deployment": {
"protocol": "A2A"
}
}This JSON is an explanatory example, not a guaranteed production-valid OASF document. Developers must validate records against the current OASF schema before deployment.
Structured taxonomies make capability-based search more useful than a human-written service catalog. However, taxonomies and extensions introduce their own compatibility questions. An extension can carry valuable organization-specific metadata, but custom fields can reduce interoperability when consumers do not understand them.
What does the AGNTCY Agent Directory do?
The Agent Directory publishes, stores and discovers OASF agent records. The Directory repository describes APIs and tools for finding records by capabilities or constraints, as well as mechanisms related to integrity and provenance.
The practical distinction is important:
- A static catalog is a centrally maintained list of known services.
- A Directory is intended to support a network-oriented discovery model with local and remote storage possibilities.
- Capability search selects possible candidates from structured skills and domains.
- Identity and provenance help determine who published or controls a record.
Discovery is candidate selection, not automatic trust. An agent can match a requested “financial document analysis” skill but lack the required authentication method, data access, region, latency, compliance certification, tool permissions or availability guarantee. A client should inspect protocol and policy metadata before attempting an invocation.
Recommended Free Tools
Records can also become stale after a model change, tool removal, API migration, permission change or taxonomy update. A production Directory should therefore have record ownership, versioning, expiration or refresh policies, and a process for revoking inaccurate records.
How can developers try the Directory locally?
The Directory quickstart provides a local developer demonstration. The current Directory getting-started documentation gives these macOS installation commands:
brew tap agntcy/dir https://github.com/agntcy/dir
brew trust agntcy/dir
brew install dirctlOn Linux, the documented platform-specific binary installation is:
curl -L https://github.com/agntcy/dir/releases/latest/download/dirctl-linux-amd64 -o dirctl
chmod +x dirctl
sudo mv dirctl /usr/local/bin/After installing dirctl, initialize and start a local node:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsdirctl init
dirctl daemon startThe quickstart documents these local services:
| Service | Local address |
|---|---|
| HTTP API and dashboard | localhost:8889 |
| gRPC Directory API | localhost:8888 |
| DHT API | localhost:8999 |
| OCI registry | localhost:5555 |
These ports and commands are version-sensitive. The Linux command uses the volatile releases/latest path, so teams writing reproducible build instructions should pin and verify a release instead of copying that URL unchanged.
dirctl init provisions an approximately 89 MB local model and taxonomy for local record enrichment and natural-language search, according to the quickstart. A local node does not automatically create a globally populated marketplace; the documentation says records remain on the local peer unless the environment is configured for broader exchange.
The Directory repository also documents PostgreSQL and remote OCI registry options, along with Docker Compose and Helm deployment paths. One repository example uses Helm chart version v1.6.2:
Also popular on Amazon
- GIGABIT ETHERNET PORTS: Features 5 x 1.0Gbps Ethernet ports for high-speed connectivity. Auto-negotiating ports detect the optimal speed for connected devices and work with existing Cat5e or Cat6 Ethernet cables.
- EASY SMART MANAGED NETWORK SWITCH: Intuitive software interface offers Easy Smart Managed Essentials capabilities to configure VLANs, prioritize traffic with QoS, monitor ports, and manage network security for small businesses.
- FLEXIBLE MOUNTING OPTIONS: Compact metal design supports desktop or wall-mount placement for versatile installation.
- SILENT & ENERGY-EFFICIENT OPERATION: Fanless design ensures silent performance, while IEEE 802.3az Energy Efficient Ethernet reduces power consumption without compromising high-speed network performance.
- REGIONAL COMPATIBILITY: Made for use in U.S. & CA only
helm pull oci://ghcr.io/agntcy/dir/helm-charts/dir --version v1.6.2
helm upgrade --install dir oci://ghcr.io/agntcy/dir/helm-charts/dir --version v1.6.2v1.6.2 is a version snapshot documented in the repository, not a claim that it is the latest release. Confirm the current chart and compatibility requirements before using the command in production.
What is ACP, and is ACP a stable standard?
ACP, the Agent Connect Protocol, was proposed as a REST/OpenAPI interface for configuring and invoking remote agents. The design aims to let agents built by different parties, using different multi-agent systems and running in different locations, interact through a common interface. The ACP specification repository and AGNTCY connection documentation describe that invocation model.
ACP should not be presented as a stable, universally adopted standard. The original agntcy/acp-spec repository was archived on April 11, 2026. The ACP SDK repository remains public and contains client and server tooling, but a public SDK demonstrates implementation availability, not broad industry adoption.
The archival event does not prove that AGNTCY as a whole was abandoned. AGNTCY continues to maintain and present components such as OASF, Directory, Identity and SLIM. The archival event does prove that individual specifications can be reorganized, superseded or paused. Teams considering ACP should check current documentation, inspect supported versions and run compatibility tests before making ACP a production dependency.
How does AGNTCY compare with LangChain, LangGraph, MCP and A2A?
AGNTCY operates at a different layer from agent-building frameworks. LangChain is positioned as a framework for building agent applications, with provider-agnostic design, tool-calling architecture and middleware. LangGraph focuses on more controlled, stateful agent workflows and orchestration, as described in LangChain’s LangChain and LangGraph 1.0 overview.
A developer can build an agent in LangGraph, LangChain, LlamaIndex, CrewAI or a custom Python service, then expose an AGNTCY-compatible record or protocol. AGNTCY is therefore not a direct LangChain competitor. AGNTCY is closer to an interoperability and platform-infrastructure layer around agents.
| Technology | Main job | What it does not solve by itself |
|---|---|---|
| AGNTCY/OASF | Describe agent capabilities and metadata. | Agent reasoning, model quality or runtime execution. |
| AGNTCY Directory | Publish and discover agents. | Whether a discovered agent is trustworthy or effective in practice. |
| AGNTCY ACP | Proposed remote-agent invocation and configuration interface. | Universal adoption or long-term specification stability. |
| AGNTCY Identity | Establish agent and multi-agent-system identities. | Correctness, authorization policy or behavioral safety by itself. |
| MCP | Connect models or agents to tools and data sources. | Complete agent-to-agent discovery and governance. |
| A2A | Support agent-to-agent communication and task exchange. | Complete discovery, observability and security governance. |
| LangChain | Build agent applications and tool-calling logic. | A neutral, cross-vendor agent registry. |
| LangGraph | Orchestrate controlled, stateful agent workflows. | Universal interoperability across runtimes. |
| LangSmith | Trace, evaluate and monitor LangChain-based applications. | A neutral communication or identity standard. |
AGNTCY does not replace MCP or A2A. MCP is generally useful for connecting an agent or model to tools and data. A2A is relevant when agents exchange tasks and results. OASF and the Directory address description and discovery, while identity, messaging and observability address additional platform concerns. AGNTCY’s OASF materials reference multiple communication approaches, including MCP and A2A-related deployment information.
The boundaries are not permanent. The ecosystem is evolving, and an end-to-end system may use LangGraph internally, MCP for tools, A2A for agent communication, OASF for descriptions and a Directory for discovery. Actual interoperability still depends on adapters, supported versions, authentication, streaming behavior and error semantics.
What role do Cisco and Galileo play?
Cisco supplied the original enterprise-networking and infrastructure context for AGNTCY, while LangChain contributed a major agent-development perspective. Galileo’s original emphasis was AI quality, evaluation, observability and trust. Galileo argued that agents acting in real systems require measurement, auditability and reliability infrastructure, not only model and tool integrations.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteGalileo also described Agent Control as an Apache 2.0-licensed open standard for governing runtime agent behavior and identified Cisco AI Defense as a launch partner. Agent Control should be treated as a related or adjacent governance effort unless current AGNTCY documentation explicitly places it in the project’s core. Agent evaluation, runtime guardrails, observability, discovery and agent-to-agent communication are related but distinct functions.
Cisco’s acquisition of Galileo changes the commercial context. Cisco’s April 9, 2026 acquisition announcement and May 22, 2026 completion update establish the relevant dates. AGNTCY’s Linux Foundation positioning may provide a more neutral governance model than a project controlled solely by one vendor, but vendor influence and governance should still be evaluated rather than assumed away.
Also popular on Amazon
- 24-Gigabit ports provide instant large file transfers
- 9K Jumbo frame improves performance of large data transfers
- Effective network monitoring via Port Mirroring, Loop Prevention and Cable Diagnostics
- Abundant VLAN features improve network security via traffic segmentation
- IGMP Snooping optimizes multicast applications
Is AGNTCY open source and free to use?
The cited AGNTCY technical repositories are publicly available and use Apache 2.0 licensing, but open-source availability is not the same as a free managed service or a turnkey enterprise platform. The AGNTCY organization, OASF repository and Directory repository provide code that teams can inspect, run and adapt.
Operating an enterprise agent interoperability layer still creates costs for infrastructure, identity management, registry administration, security review, policy enforcement, monitoring, upgrades, compatibility testing and integration work. A project can be open source while its surrounding hosted support, security products or enterprise guarantees are commercial.
What are the main security and reliability risks?
AGNTCY can supply useful building blocks for trust, but no directory or identity mechanism makes an agent automatically safe.
- Impersonation: An attacker may publish a record that resembles a trusted internal or vendor agent. Identity, signatures and provenance can help establish who published a record, but they do not prove that the agent is competent.
- Poisoned metadata: A malicious or inaccurate record may overclaim skills, permissions, compliance status or availability.
- Stale records: A previously valid description may remain visible after an API, model, tool or permission change.
- Protocol mismatch: An agent may be discoverable through OASF but unreachable because the consumer does not support the listed deployment or communication protocol.
- Framework leakage: A supposedly portable agent may expose LangGraph state concepts, proprietary authentication, vendor-specific tool schemas, nonportable streaming behavior or framework-specific error semantics.
- Distributed-system complexity: Remote discovery and invocation add network failures, retries, timeouts, partial results and authorization boundaries.
Cryptographic identity answers “who published or controls this identity?” It does not answer “will this agent produce a correct answer?” Authorization answers “may this caller perform this action?” It does not answer “should the action be allowed for this particular task?” Runtime guardrails, evaluation, audit logs and human approval may still be required.
Who should use AGNTCY?
AGNTCY is worth evaluating when interoperability is more important than the shortest path to a single-framework prototype.
| Situation | AGNTCY fit | Reason |
|---|---|---|
| Several internal teams publish agents for reuse. | Strong candidate | Structured records and discovery can reduce manually maintained endpoints and catalogs. |
| Agents are built with different frameworks. | Strong candidate for experimentation | The project’s central ambition is framework-agnostic description and connection. |
| Agents operate across organizations or environments. | Potentially valuable | Identity, provenance, protocol metadata and discovery become more important at organizational boundaries. |
| One agent needs access to tools and data. | Usually start with MCP | A simple MCP integration may solve tool access without introducing a directory or multi-agent layer. |
| One application contains a controlled workflow. | Usually start with LangGraph or another orchestrator | A local workflow does not need cross-organization discovery merely because it contains multiple steps or agents. |
| The main need is tracing or evaluation. | Use an observability or evaluation tool | Discovery and schemas do not replace application traces, quality tests or runtime monitoring. |
| The organization needs a stable, widely adopted standard immediately. | High caution | Specifications are evolving, ACP’s original repository was archived and independent adoption evidence is limited in the supplied sources. |
AGNTCY is probably unnecessary for a single-agent proof of concept, a small application contained inside one LangGraph or LangChain deployment, or a team that cannot maintain custom registries and evolving integrations.
What should a technical team test before production?
A team should treat AGNTCY as an integration surface to validate, not as a badge that establishes production readiness.
- Define the boundary: Decide whether the immediate problem is agent construction, tool access, agent discovery, cross-agent communication, identity, observability or governance.
- Describe one real agent: Create an OASF record for a narrow capability and compare the record with the agent’s actual tools, data access, protocols and permissions.
- Run a local Directory: Use the documented quickstart to publish and find records locally before designing a distributed marketplace.
- Test negative cases: Try stale metadata, missing authentication, unsupported protocol, unavailable agent, timeout, malformed result and revoked identity.
- Separate discovery from authorization: Require explicit policy approval before a discovered agent can receive sensitive data or invoke consequential tools.
- Measure behavioral interoperability: Test streaming, state, retries, error semantics, provenance and result formats. A shared schema does not guarantee that two agents behave compatibly.
- Pin versions: Avoid relying on floating release URLs or unverified chart versions. Check current repositories and documentation before deployment.
- Plan for change: Define how records expire, how protocol versions are negotiated, how agents are revoked and how consumers react when a specification changes.
Where can teams add commercial tooling?
AGNTCY itself is an open-source project, so the natural commercial layer is adjacent tooling rather than a required AGNTCY subscription. A LangChain or LangGraph team may evaluate LangSmith for tracing, debugging, evaluation and monitoring. Those capabilities address a different problem from agent discovery, but they become useful when a multi-agent system needs evidence about what happened during an interaction.
Enterprise teams assessing runtime controls can review Cisco AI Defense and the related Agent Control material. Cisco products are not required to use AGNTCY, and open AGNTCY repositories do not automatically provide the managed security controls or enterprise support associated with a commercial platform.
Organizations operating distributed infrastructure may also compare broader application and infrastructure observability products such as Splunk Observability Cloud. The right choice depends on whether the team needs model traces, agent evaluations, runtime policy, service telemetry or all of those layers.
What is AGNTCY’s maturity level in 2026?
AGNTCY is active and usable for technical evaluation, but the supplied evidence does not establish that AGNTCY is a settled universal industry standard or broadly adopted production platform. Public repositories, quickstarts, schemas and governance documentation demonstrate project activity. They do not establish market share, independent production scale, universal compatibility or benchmark superiority.
Also popular on Amazon
- 16 10/100/1000Mbps RJ45 Ports
- Plug and play, with No configuration required
- Durable metal casing of superior quality and Professional appearance
- Intelligent management via a web user interface and downloadable Utility
- Green technology reduces power consumption
The archived ACP specification is a particularly important maturity signal. It does not invalidate the broader project, but it shows why teams should evaluate each component independently. OASF, Directory, Identity and SLIM may have different release histories, compatibility guarantees and adoption levels.
Open source, Apache 2.0 licensing, Linux Foundation positioning and public governance can improve inspectability and reduce dependence on one vendor. None of those properties alone guarantees stable APIs, secure defaults, reliable agent behavior or a large implementation community.
Bottom line: should developers use AGNTCY?
Developers should experiment with AGNTCY when they are building a multi-agent platform that spans frameworks, teams or organizations and needs structured discovery, identity and interoperability. Developers should not adopt AGNTCY simply because they need to build one agent, connect a model to a few tools or orchestrate a workflow inside one application.
The project’s strongest idea is separation of concerns: build an agent with the framework that suits the application, then use shared infrastructure to describe and connect that agent to others. The project’s main risks are changing specifications, integration overhead, stale or misleading metadata, security boundaries and uncertain independent adoption.
In practical terms, AGNTCY is worth watching and prototyping, but it should not yet be described as a finished universal replacement for LangChain, LangGraph, MCP, A2A or observability platforms.
Frequently Asked Questions
Is AGNTCY a replacement for LangChain or LangGraph?
No. AGNTCY is an interoperability and infrastructure project for describing, discovering, identifying and connecting agents, while LangChain and LangGraph are used to build and orchestrate agent applications. A LangGraph agent could expose AGNTCY-compatible metadata or interfaces.
Does AGNTCY replace MCP or A2A?
No. MCP primarily connects models or agents to tools and data, while A2A focuses on agent-to-agent communication and task exchange. AGNTCY addresses additional layers such as agent descriptions, discovery, identity and supporting infrastructure, and its materials reference integration with multiple protocols.
Can I use AGNTCY in production today?
Teams can evaluate and deploy individual AGNTCY components, but production adoption requires component-by-component compatibility and security testing. The original ACP specification repository was archived on April 11, 2026, so teams should verify current specifications, versions and documentation before depending on ACP.
Does finding an agent in the AGNTCY Directory mean that the agent is trustworthy?
No. Directory discovery identifies a candidate based on published metadata. Identity and provenance can help establish who published a record, but teams still need authorization, security controls, freshness checks, behavioral evaluation and policy review.
The Bottom Line
AGNTCY is an open interoperability layer for multi-agent systems, not a finished universal agent-building framework. Use it when cross-framework discovery and communication are genuine requirements; otherwise, begin with the simplest tool or agent framework that solves the immediate problem.
