# The AI Agent Stack: Architecture Layers and Market Map

**Research date:** June 15, 2026  
**Scope:** Active or influential open-source projects and commercial products with a public official website or repository.

![The AI Agent Application Stack](ai-agent-application-stack-landscape-optimized.webp)

## How to read this map

There is no single “AI agent market.” There are several markets stacked on top of one another:

```text
Agent products and hosts
User experience and agent interaction
Application and model SDKs
Agent frameworks and orchestration
Managed agent infrastructure and durable execution
Tools, integrations, protocols, retrieval, and memory
Sandboxed execution and computer use
Model gateways and routing
Model providers and inference runtimes
Compute, storage, and network
```

Security, identity, governance, observability, evaluation, and cost management cut across the entire stack.

Products often span several layers. This directory places each player where an architect would primarily evaluate or buy it. “Comprehensive” here means a broad, useful working set, not every GitHub repository or newly launched startup.

## Layer 0: Agent products and hosts

This is the layer users operate directly. These products contain or connect to an agent runtime, but they are evaluated as finished coding agents, desktop agents, IDEs, assistants, or automation products.

### Open-source and model-flexible coding agents

| Player | Primary surface | What it is trying to solve |
|---|---|---|
| [goose](https://block.github.io/goose/) | Desktop, CLI, API | Block's open-source, general-purpose local agent for coding, automation, research, workflows, MCP extensions, and multi-agent delegation |
| [OpenCode](https://opencode.ai/) | Terminal, desktop, IDE | Provider-neutral open-source coding agent with a client/server architecture, LSP awareness, tools, permissions, agents, and MCP |
| [Cline](https://cline.bot/) | VS Code, CLI, SDK | Open coding-agent runtime with plan/action workflows, model choice, MCP, and embeddability |
| [Roo Code](https://roocode.com/) | VS Code | Open-source coding agent focused on configurable modes, tools, model choice, and extension workflows |
| [Kilo Code](https://kilo.ai/) | IDE, CLI, cloud | Open-source coding-agent platform spanning local development, terminal use, cloud agents, and team workflows |
| [Continue](https://www.continue.dev/) | VS Code, JetBrains, CLI | Open-source coding-assistant platform for custom models, rules, context providers, autocomplete, and agents |
| [Aider](https://aider.chat/) | Terminal | Git-native pair-programming agent with broad model support and repository map/context management |
| [OpenHands](https://openhands.dev/) | Web, CLI, SDK, cloud | Open platform for software-development agents, remote execution, evaluation, and delegation |
| [SWE-agent](https://swe-agent.com/) | CLI, research framework | Open software-engineering agent and benchmark-oriented environment from Princeton |
| [Qwen Code](https://github.com/QwenLM/qwen-code) | Terminal | Open-source terminal coding agent optimized for Qwen but usable with compatible model endpoints |
| [Gemini CLI](https://github.com/google-gemini/gemini-cli) | Terminal | Google's open-source agentic terminal interface for Gemini and developer workflows |
| [OpenAI Codex](https://openai.com/codex/) | CLI, IDE, desktop, cloud | OpenAI's coding agent for local interactive work and delegated cloud tasks |
| [Tabby](https://tabbyml.com/) | IDE and self-hosted server | Self-hosted open-source coding assistant and completion platform |
| [Zed](https://zed.dev/ai) | Native editor | High-performance editor with first-party and Agent Client Protocol integrations |
| [Open Interpreter](https://openinterpreter.com/) | Terminal and local computer | Open agent that lets language models operate code and the user's computer |
| [Agent Zero](https://github.com/frdel/agent-zero) | Web/terminal agent | Open general-purpose agent with tools, subagents, memory, and local control |
| [AutoGPT](https://agpt.co/) | Agent platform | Open agent platform and visual system for creating and running autonomous workflows |
| [Browser Use](https://browser-use.com/) | Library, CLI, cloud agent | Browser-operating agents and cloud browser automation |

### Commercial and vendor-native coding agents

| Player | Primary surface | Center of gravity |
|---|---|---|
| [Claude Code](https://www.anthropic.com/claude-code) | Terminal, IDE, web | Anthropic-native coding agent and agent SDK ecosystem |
| [GitHub Copilot](https://github.com/features/copilot) | IDE, CLI, GitHub, cloud | Coding assistance, repository agents, reviews, issue-to-PR work, and multi-agent hosting |
| [Cursor](https://cursor.com/) | AI-native editor and cloud agents | Integrated coding, codebase reasoning, background agents, review, and team workflows |
| [Windsurf](https://windsurf.com/) | AI-native editor | Agentic coding IDE with codebase context and workflow automation |
| [Devin](https://devin.ai/) | Cloud software engineer | Delegated autonomous software-development tasks in managed environments |
| [Amazon Q Developer](https://aws.amazon.com/q/developer/) | IDE, CLI, AWS console | AWS-oriented development, transformation, operations, and cloud assistance |
| [JetBrains Junie](https://www.jetbrains.com/junie/) | JetBrains IDEs | Coding agent integrated with JetBrains project intelligence |
| [Warp](https://www.warp.dev/) | Agentic terminal | Terminal-native coding and DevOps agents with shared workflows |
| [Amp](https://ampcode.com/) | Terminal and editor | Sourcegraph's agentic coding product for large codebases and delegated work |
| [Replit Agent](https://replit.com/ai) | Browser IDE and hosted runtime | Prompt-to-application building, execution, and deployment |
| [Bolt](https://bolt.new/) | Browser app builder | Prompt-to-full-stack web application generation and deployment |
| [Lovable](https://lovable.dev/) | Browser app builder | Conversational application generation with visual editing and deployment |
| [v0](https://v0.dev/) | Browser and Vercel ecosystem | Generative UI and full-stack web application creation |
| [Firebase Studio](https://firebase.studio/) | Browser development environment | Google/Firebase-oriented agentic application building |
| [Mistral Vibe](https://mistral.ai/products/vibe) | Coding agent | Mistral-native agentic software-development experience |

### Business and knowledge-work agent products

| Player | Primary use |
|---|---|
| [Glean Agents](https://www.glean.com/product/agents) | Enterprise search, knowledge, and workplace agents |
| [Salesforce Agentforce](https://www.salesforce.com/agentforce/) | CRM, customer service, sales, and enterprise workflow agents |
| [Microsoft Copilot Studio](https://www.microsoft.com/microsoft-copilot/microsoft-copilot-studio) | Enterprise copilots and low-code agents across Microsoft systems |
| [Google Agentspace](https://cloud.google.com/products/agentspace) | Enterprise search and agent experiences over organizational data |
| [ServiceNow AI Agents](https://www.servicenow.com/products/ai-agents.html) | IT, employee, customer-service, and workflow agents |
| [UiPath Agentic Automation](https://www.uipath.com/platform/agentic-automation) | Agents combined with robotic process automation and enterprise workflows |
| [Dust](https://dust.tt/) | Custom enterprise assistants grounded in company systems |
| [Lindy](https://www.lindy.ai/) | No-code business and personal workflow agents |
| [Relevance AI](https://relevanceai.com/) | No-code agent teams and business automation |
| [Gumloop](https://www.gumloop.com/) | Visual AI workflow and automation agents |
| [Relay.app](https://www.relay.app/) | Human-in-the-loop workflow automation |
| [Harvey](https://www.harvey.ai/) | Legal and professional-services agents |
| [Sierra](https://sierra.ai/) | Customer-experience agents |
| [Intercom Fin](https://www.intercom.com/fin) | Customer-support agent |

## Layer 1: User experience and agent interaction

This layer turns backend agents into usable products. It handles chat, multimodal input, streamed events, generative UI, frontend tools, shared application state, approvals, and human intervention.

### Agentic frontend frameworks and component systems

| Player | Primary fit |
|---|---|
| [CopilotKit](https://www.copilotkit.ai/) | Embedded copilots, AG-UI, generative UI, frontend tools, shared state, and human-in-the-loop interaction |
| [assistant-ui](https://www.assistant-ui.com/) | Composable React chat and agent interfaces with runtime adapters |
| [AI Elements](https://ai-sdk.dev/elements) | Vercel/shadcn-style components for AI application interfaces |
| [Chainlit](https://chainlit.io/) | Python-first conversational and agent application UI |
| [Gradio](https://www.gradio.app/) | Rapid Python interfaces for models, multimodal apps, and agents |
| [Streamlit](https://streamlit.io/) | Python data and AI application interfaces |
| [NLUX](https://nlux.ai/) | Framework-neutral conversational AI UI library |
| [Stream Chat](https://getstream.io/chat/) | Production chat infrastructure and AI-agent messaging interfaces |
| [Botpress Webchat](https://botpress.com/docs/webchat) | Embeddable conversational-agent UI |
| [Voiceflow](https://www.voiceflow.com/) | Conversational and voice-agent design and prototyping |

### Self-hosted assistant shells

| Player | Primary fit |
|---|---|
| [Open WebUI](https://openwebui.com/) | Self-hosted multi-model and local-model interface |
| [LibreChat](https://www.librechat.ai/) | Self-hosted multi-provider assistant platform |
| [Chatbot UI](https://www.chatbotui.com/) | Open-source conversational UI |
| [LobeChat](https://lobehub.com/) | Open-source multi-provider chat and agent workspace |
| [AnythingLLM](https://anythingllm.com/) | Self-hosted chat, RAG, workspaces, and agents |

### Voice and realtime-agent interfaces

| Player | Primary fit |
|---|---|
| [LiveKit Agents](https://docs.livekit.io/agents/) | Realtime voice, video, and multimodal agent framework |
| [Vapi](https://vapi.ai/) | Voice-agent development and telephony platform |
| [Retell AI](https://www.retellai.com/) | Production phone and voice agents |
| [Bland AI](https://www.bland.ai/) | Phone-call automation agents |
| [Pipecat](https://pipecat.ai/) | Open-source realtime voice and multimodal agent framework |
| [Daily Bots](https://www.daily.co/products/daily-bots/) | Realtime voice/video agent infrastructure |

## Layer 2: Interaction and interoperability protocols

Protocols are not agent frameworks. They standardize boundaries between hosts, tools, agents, user interfaces, commerce systems, and external services.

| Protocol | Boundary it standardizes |
|---|---|
| [Model Context Protocol (MCP)](https://modelcontextprotocol.io/) | Agent/host access to tools, prompts, resources, and external context |
| [Agent2Agent Protocol (A2A)](https://a2a-protocol.org/) | Discovery, delegation, messaging, and task lifecycle between agents |
| [AG-UI](https://ag-ui.com/) | Event stream between agent backends and user-facing applications |
| [Agent Client Protocol (ACP)](https://agentclientprotocol.com/) | Communication between coding agents and editors such as Zed |
| [A2UI](https://a2ui.org/) | Declarative agent-generated user interfaces |
| [MCP Apps](https://github.com/modelcontextprotocol/ext-apps) | Interactive UI resources delivered through MCP hosts |
| [Agent Payments Protocol (AP2)](https://ap2-protocol.org/) | Agent-initiated payments and commerce authorization |
| [Universal Commerce Protocol (UCP)](https://ucp.dev/) | Interoperable commerce flows for agents and businesses |
| [Agent Network Protocol (ANP)](https://agent-network-protocol.com/) | Decentralized agent discovery, identity, and communication |
| [OpenAPI](https://www.openapis.org/) | Machine-readable HTTP APIs commonly converted into agent tools |
| [JSON Schema](https://json-schema.org/) | Structured tool inputs, outputs, and generated data |
| [OpenTelemetry](https://opentelemetry.io/) | Vendor-neutral traces, metrics, and logs, including emerging GenAI conventions |

## Layer 3: AI application and model SDKs

These libraries normalize model access and application primitives such as streaming, structured output, multimodal input, tools, prompt management, and frontend state. Some overlap with the agent-runtime layer.

| Player | Language/ecosystem | Center of gravity |
|---|---|---|
| [Vercel AI SDK](https://ai-sdk.dev/) | TypeScript | Full-stack AI applications, provider abstraction, streaming, tools, agents, and UI |
| [TanStack AI](https://tanstack.com/ai/latest) | TypeScript | Modular, type-safe provider adapters, tools, streaming, and application clients |
| [Genkit](https://genkit.dev/) | JavaScript/TypeScript, Go | Full-stack AI flows, tools, evaluation, local development, and Google integrations |
| [LangChain](https://www.langchain.com/langchain) | Python, JavaScript | Broad component ecosystem for models, tools, retrieval, agents, and chains |
| [LlamaIndex](https://www.llamaindex.ai/) | Python, TypeScript | Data-centric AI applications, retrieval, workflows, and agents |
| [Haystack](https://haystack.deepset.ai/) | Python | Production RAG pipelines, components, tools, and agents |
| [Spring AI](https://spring.io/projects/spring-ai) | Java | Spring-native model, vector-store, tool, RAG, and agent application APIs |
| [Semantic Kernel](https://learn.microsoft.com/semantic-kernel/) | .NET, Python, Java | Enterprise application SDK for models, plugins, planning, and agent integration |
| [Instructor](https://python.useinstructor.com/) | Python and ports | Reliable schema-validated structured extraction from model outputs |
| [BAML](https://boundaryml.com/) | Multi-language | Typed prompt functions and structured model outputs |
| [Outlines](https://dottxt-ai.github.io/outlines/) | Python | Constrained and structured generation |
| [Guidance](https://github.com/guidance-ai/guidance) | Python | Controlled generation and prompt programming |
| [LMQL](https://lmql.ai/) | Python/query language | Declarative constraints and programming over language models |
| [DSPy](https://dspy.ai/) | Python | Declarative LM programs and prompt/module optimization |

## Layer 4: Agent frameworks, runtimes, and orchestration

This is the decision-loop layer. It coordinates models, tools, state, delegation, retries, approvals, and multi-agent workflows.

### Code-first agent frameworks

| Player | Primary design |
|---|---|
| [Strands Agents SDK](https://strandsagents.com/) | Model-driven agents, MCP/A2A, multi-agent Graph/Swarm, Python and TypeScript |
| [OpenAI Agents SDK](https://openai.github.io/openai-agents-python/) | Agents, handoffs, guardrails, sessions, tracing, MCP, voice, and OpenAI-centric workflows |
| [Claude Agent SDK](https://docs.anthropic.com/en/docs/claude-code/sdk) | Programmable access to Claude Code's agent harness and tools |
| [Google Agent Development Kit](https://google.github.io/adk-docs/) | Model-flexible agent development, multi-agent composition, sessions, and deployment |
| [Microsoft Agent Framework](https://learn.microsoft.com/agent-framework/) | Microsoft's successor path combining agent orchestration concepts from AutoGen and Semantic Kernel |
| [LangGraph](https://www.langchain.com/langgraph) | Stateful graphs, durable execution, interrupts, replay, and human-in-the-loop agents |
| [CrewAI](https://www.crewai.com/) | Role-based agent teams, crews, flows, and managed deployment |
| [PydanticAI](https://ai.pydantic.dev/) | Type-safe Python agents with Pydantic validation and dependency injection |
| [Mastra](https://mastra.ai/) | TypeScript agents, workflows, memory, RAG, evaluation, and development tooling |
| [Agno](https://www.agno.com/) | Python agent framework with tools, teams, memory, knowledge, and runtime |
| [smolagents](https://huggingface.co/docs/smolagents/) | Lightweight Hugging Face agents and code agents |
| [LlamaIndex Workflows](https://developers.llamaindex.ai/python/workflows/) | Event-driven workflows and data-aware agents |
| [AutoGen](https://microsoft.github.io/autogen/) | Microsoft's established multi-agent framework; increasingly complemented by Microsoft Agent Framework |
| [AG2](https://ag2.ai/) | Community-led continuation of the AutoGen multi-agent ecosystem |
| [BeeAI Framework](https://framework.beeai.dev/) | IBM-backed open-source agents and multi-agent workflows |
| [Letta](https://www.letta.com/) | Stateful agents with explicit memory architecture and long-running identity |
| [Langroid](https://langroid.github.io/langroid/) | Multi-agent Python programming framework |
| [CAMEL-AI](https://www.camel-ai.org/) | Multi-agent research and framework ecosystem |
| [Atomic Agents](https://github.com/BrainBlend-AI/atomic-agents) | Small, composable, schema-driven Python agents |
| [VoltAgent](https://voltagent.dev/) | Open-source TypeScript agent framework and observability ecosystem |
| [mcp-agent](https://github.com/lastmile-ai/mcp-agent) | Lightweight agent workflows centered on MCP |
| [Rasa](https://rasa.com/) | Enterprise conversational AI and controlled agent workflows |

### Visual and low-code agent builders

| Player | Primary fit |
|---|---|
| [Dify](https://dify.ai/) | Open-source visual AI application, RAG, workflow, and agent platform |
| [Flowise](https://flowiseai.com/) | Open-source visual agent and LLM workflow builder |
| [Langflow](https://www.langflow.org/) | Python-based visual builder for agents, MCP, and RAG |
| [n8n](https://n8n.io/) | Workflow automation with AI agents and broad integrations |
| [Botpress](https://botpress.com/) | Visual conversational and customer-facing agent platform |
| [Microsoft Copilot Studio](https://www.microsoft.com/microsoft-copilot/microsoft-copilot-studio) | Enterprise low-code agents in the Microsoft ecosystem |
| [Vertex AI Agent Builder](https://cloud.google.com/products/agent-builder) | Google's managed agent and enterprise-search builder |
| [Amazon Bedrock Agents](https://aws.amazon.com/bedrock/agents/) | AWS-managed agent building, action groups, knowledge bases, and multi-agent collaboration |
| [Vellum](https://www.vellum.ai/) | Visual AI workflow, prompt, evaluation, and deployment platform |
| [Stack AI](https://www.stack-ai.com/) | Enterprise no-code agent and workflow platform |
| [Retool Agents](https://retool.com/ai) | Agents connected to internal tools, databases, and applications |
| [Zapier Agents](https://zapier.com/agents) | Agents over Zapier's application and automation ecosystem |

## Layer 5: Managed agent infrastructure and durable execution

This layer runs and operates agents in production. It provides hosting, session isolation, identity, scaling, durable state, scheduling, deployment, and operational controls.

### Agent-native managed platforms

| Player | Primary fit |
|---|---|
| [Amazon Bedrock AgentCore](https://aws.amazon.com/bedrock/agentcore/) | Framework-neutral AWS runtime, memory, gateway, identity, policy, observability, browser, code interpreter, registry, and evaluations |
| [Microsoft Foundry Agent Service](https://learn.microsoft.com/azure/foundry/agents/overview) | Managed build, deployment, tools, tracing, identity, and scaling for agents on Azure |
| [Vertex AI Agent Engine](https://cloud.google.com/vertex-ai/generative-ai/docs/agent-engine/overview) | Managed runtime, sessions, memory, evaluation, and deployment on Google Cloud |
| [Cloudflare Agents](https://developers.cloudflare.com/agents/) | Stateful edge agents using Workers, Durable Objects, realtime connections, and workflows |
| [LangSmith Deployment](https://www.langchain.com/langsmith/deployment) | Managed deployment and operations for LangGraph applications |
| [CrewAI AMP](https://www.crewai.com/) | Managed enterprise deployment and operation of CrewAI agents |
| [Mastra Cloud](https://mastra.ai/cloud) | Managed deployment for Mastra applications and agents |
| [Agentuity](https://agentuity.com/) | Agent-native cloud, development, deployment, routing, and observability |
| [Blaxel](https://blaxel.ai/) | Serverless agent infrastructure, sandboxes, functions, and model endpoints |
| [xpander.ai](https://xpander.ai/) | Full-lifecycle, framework-neutral agent platform and control plane |
| [Inkeep](https://inkeep.com/) | Agent builder, platform, and customer-facing agent deployment |

### Durable workflow and background execution platforms

| Player | Primary fit |
|---|---|
| [Temporal](https://temporal.io/) | Highly durable, replayable, long-running workflows |
| [Inngest](https://www.inngest.com/) | Event-driven durable functions and agent workflows |
| [Trigger.dev](https://trigger.dev/) | Long-running background jobs and AI tasks |
| [Hatchet](https://hatchet.run/) | Open-source durable task and workflow orchestration |
| [Restate](https://restate.dev/) | Durable execution, virtual objects, and workflows |
| [DBOS](https://www.dbos.dev/) | Database-backed durable workflows and application execution |
| [Prefect](https://www.prefect.io/) | Python workflow orchestration and data/AI pipelines |
| [Dagster](https://dagster.io/) | Data and AI asset orchestration |
| [Apache Airflow](https://airflow.apache.org/) | Scheduled batch workflow orchestration |

## Layer 6: Tools, actions, integrations, and web intelligence

Agents need authenticated, governed access to business applications, APIs, browsers, search, and the public web.

### Tool and integration platforms

| Player | Primary fit |
|---|---|
| [Composio](https://composio.dev/) | Agent toolkits, managed authentication, and broad SaaS integrations |
| [Arcade](https://www.arcade.dev/) | Secure agent tools, user authorization, MCP, and tool evaluation |
| [Pipedream](https://pipedream.com/) | Developer integration platform, workflows, and MCP-connected APIs |
| [Nango](https://nango.dev/) | OAuth, credentials, API integrations, and agent connectivity |
| [Klavis AI](https://www.klavis.ai/) | Hosted MCP infrastructure, tool discovery, and enterprise integration |
| [Zapier MCP](https://zapier.com/mcp) | Agent access to Zapier's application ecosystem |
| [Apify](https://apify.com/ai-agents) | Web scraping, browser automation, Actors, and agent tools |
| [Activepieces](https://www.activepieces.com/) | Open-source automation and MCP-enabled integrations |
| [Paragon](https://www.useparagon.com/) | Embedded integrations and managed authentication |
| [Merge](https://www.merge.dev/) | Unified APIs and agent integrations for business software |
| [Workato](https://www.workato.com/) | Enterprise integration, automation, and agent orchestration |
| [Tray.ai](https://tray.ai/) | Enterprise integration and agent connectivity |
| [Make](https://www.make.com/) | Visual automation and application integrations |
| [Toolhouse](https://toolhouse.ai/) | Tool infrastructure, memory, and agent capabilities |
| [Smithery](https://smithery.ai/) | MCP server registry, hosting, and discovery |
| [Glama](https://glama.ai/mcp/servers) | MCP server directory and hosting ecosystem |

### Search, research, crawling, and browser data

| Player | Primary fit |
|---|---|
| [Tavily](https://www.tavily.com/) | Search and research API designed for agents |
| [Exa](https://exa.ai/) | Neural web search and research APIs |
| [Firecrawl](https://www.firecrawl.dev/) | Website crawling, extraction, search, and agent-ready content |
| [Jina AI](https://jina.ai/) | Search, readers, rerankers, embeddings, and web data APIs |
| [Parallel Web Systems](https://parallel.ai/) | Web research and data retrieval infrastructure for agents |
| [Perplexity Sonar](https://docs.perplexity.ai/) | Search-grounded answer and research APIs |
| [SerpAPI](https://serpapi.com/) | Structured search-engine result APIs |
| [Bright Data](https://brightdata.com/ai) | Web data, scraping, proxies, and agent infrastructure |
| [Oxylabs AI Studio](https://aistudio.oxylabs.io/) | Web intelligence, crawling, extraction, search, and browser-agent infrastructure |
| [Diffbot](https://www.diffbot.com/) | Knowledge graph and structured web extraction |

## Layer 7: Sandboxed execution and agent computers

This layer gives agents isolated machines, filesystems, processes, browsers, desktops, code interpreters, and preview environments.

### General code and computer sandboxes

| Player | Primary fit |
|---|---|
| [Daytona](https://www.daytona.io/) | Framework-neutral programmable computers, snapshots, code interpreter, Git, previews, GPUs, Windows, and computer use |
| [E2B](https://e2b.dev/) | Hosted secure cloud sandboxes and code interpreters for agents |
| [Runloop](https://runloop.ai/) | Development environments and infrastructure for coding agents |
| [Modal Sandboxes](https://modal.com/docs/guide/sandbox) | On-demand containers and sandboxes within a serverless compute platform |
| [Vercel Sandbox](https://vercel.com/docs/vercel-sandbox) | Firecracker-based isolated execution integrated with Vercel |
| [Cloudflare Sandbox](https://developers.cloudflare.com/sandbox/) | Sandboxed code execution integrated with Workers and Containers |
| [Fly.io Sprites](https://sprites.dev/) | Persistent, fast-starting remote computers for agents |
| [Blaxel Sandboxes](https://docs.blaxel.ai/Sandboxes/Overview) | Serverless agent sandboxes and execution environments |
| [Northflank](https://northflank.com/product/sandboxes) | Container infrastructure and isolated workloads for agents |
| [Beam](https://www.beam.cloud/) | Serverless CPU/GPU execution and sandboxes |
| [Freestyle](https://www.freestyle.sh/) | Stateful agent computers and code execution |
| [CodeSandbox SDK](https://codesandbox.io/docs/sdk) | Programmatic development sandboxes and previews |
| [Replit](https://replit.com/) | Hosted development environments and application execution |
| [Azure Container Apps dynamic sessions](https://learn.microsoft.com/azure/container-apps/sessions) | Azure-managed isolated code-interpreter and custom-container sessions |
| [AgentCore Code Interpreter](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/code-interpreter.html) | AWS-managed isolated code execution |

### Browser and desktop computer-use infrastructure

| Player | Primary fit |
|---|---|
| [Browserbase](https://www.browserbase.com/) | Managed browsers, sessions, observability, and agent browser infrastructure |
| [Steel](https://steel.dev/) | Open-source browser API and managed browser infrastructure |
| [Hyperbrowser](https://hyperbrowser.ai/) | Cloud browsers and web agents |
| [Kernel](https://www.onkernel.com/) | Browser and computer infrastructure for agents |
| [Browserless](https://www.browserless.io/) | Hosted browser automation and Chrome infrastructure |
| [Browser Use Cloud](https://cloud.browser-use.com/) | Managed browser-use agents and browser sessions |
| [Anchor Browser](https://anchorbrowser.io/) | Cloud browser infrastructure for agents |
| [AgentCore Browser](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/browser-tool.html) | AWS-managed browser automation |

## Layer 8: Memory, retrieval, context, and knowledge

This layer turns stateless model calls into systems that can remember users, retrieve organizational knowledge, and maintain durable context.

### Agent memory and context engines

| Player | Primary fit |
|---|---|
| [Mem0](https://mem0.ai/) | User, session, agent, and graph memory as a framework or managed service |
| [Zep](https://www.getzep.com/) | Context graphs and enterprise agent memory |
| [Letta](https://www.letta.com/) | Stateful agents with explicit working and archival memory |
| [Cognee](https://www.cognee.ai/) | Knowledge graphs and memory for agents |
| [Graphiti](https://github.com/getzep/graphiti) | Temporal knowledge graphs for agent memory |
| [Supermemory](https://supermemory.ai/) | Universal memory and context API |
| [LangMem](https://langchain-ai.github.io/langmem/) | Long-term memory tools for LangGraph agents |
| [Hindsight](https://hindsight.vectorize.io/) | Agent memory and retrieval system |
| [EverMind](https://evermind.ai/) | Persistent agent memory platform |
| [Memvid](https://memvid.com/) | Portable file-oriented memory for AI applications |

### Retrieval and document processing frameworks

| Player | Primary fit |
|---|---|
| [LlamaIndex](https://www.llamaindex.ai/) | Data connectors, indexing, retrieval, workflows, and agents |
| [Haystack](https://haystack.deepset.ai/) | RAG pipelines, document stores, retrievers, and agents |
| [LangChain](https://www.langchain.com/) | Retrieval components, loaders, splitters, vector stores, and agents |
| [Unstructured](https://unstructured.io/) | Document ingestion, parsing, partitioning, and enrichment |
| [LlamaParse](https://www.llamaindex.ai/llamaparse) | Agentic document parsing |
| [Docling](https://docling-project.github.io/docling/) | Open document conversion and structured extraction |
| [Vectorize](https://vectorize.io/) | Managed RAG pipelines and context engineering |
| [Graphlit](https://www.graphlit.com/) | Multimodal knowledge ingestion, extraction, search, and agent memory |

### Vector and hybrid-search databases

| Player | Primary fit |
|---|---|
| [Pinecone](https://www.pinecone.io/) | Managed vector database and retrieval |
| [Weaviate](https://weaviate.io/) | Open-source and managed vector/hybrid database |
| [Qdrant](https://qdrant.tech/) | Open-source and managed vector database |
| [Milvus / Zilliz](https://milvus.io/) | Distributed open-source vector database and managed cloud |
| [Chroma](https://www.trychroma.com/) | Developer-focused open-source and hosted retrieval database |
| [LanceDB](https://lancedb.com/) | Embedded and cloud multimodal/vector database |
| [pgvector](https://github.com/pgvector/pgvector) | Vector search extension for PostgreSQL |
| [Redis](https://redis.io/solutions/vector-database/) | In-memory data platform with vector and hybrid search |
| [Elasticsearch](https://www.elastic.co/elasticsearch/vector-database) | Search, hybrid retrieval, and vector database capabilities |
| [MongoDB Atlas Vector Search](https://www.mongodb.com/products/platform/atlas-vector-search) | Vector search within MongoDB |
| [Vespa](https://vespa.ai/) | Large-scale search, ranking, recommendations, and vectors |
| [Turbopuffer](https://turbopuffer.com/) | Serverless vector and full-text search |
| [Supabase Vector](https://supabase.com/docs/guides/ai) | Postgres/pgvector retrieval in the Supabase platform |
| [OpenSearch](https://opensearch.org/platform/search/vector-database.html) | Open-source search and vector retrieval |

## Layer 9: Observability, evaluation, testing, and prompt operations

Traditional logs tell you whether code crashed. Agent observability must also explain model decisions, tool calls, trajectories, quality, cost, regressions, and user outcomes.

### End-to-end observability and evaluation platforms

| Player | Primary fit |
|---|---|
| [LangSmith](https://www.langchain.com/langsmith) | Agent tracing, datasets, evaluation, prompt management, and deployment |
| [Langfuse](https://langfuse.com/) | Open-source traces, evaluation, prompts, metrics, and self-hosting |
| [Braintrust](https://www.braintrust.dev/) | Evals, experiments, datasets, tracing, and production monitoring |
| [Arize Phoenix](https://phoenix.arize.com/) | Open-source OpenTelemetry-native tracing and evaluation |
| [Weights & Biases Weave](https://wandb.ai/site/weave) | Tracing, evaluation, datasets, and model/application analytics |
| [AgentOps](https://www.agentops.ai/) | Agent monitoring, replay, cost tracking, and evaluation |
| [Helicone](https://www.helicone.ai/) | Gateway, observability, sessions, prompts, and evaluations |
| [Opik](https://www.comet.com/site/products/opik/) | Open-source tracing, evaluation, prompt optimization, and production monitoring |
| [MLflow](https://mlflow.org/genai) | Open-source GenAI tracing, evaluation, prompt/version management, and deployment integration |
| [Traceloop](https://www.traceloop.com/) | OpenLLMetry/OpenTelemetry-based LLM and agent observability |
| [Parea](https://www.parea.ai/) | Evaluation, testing, tracing, and experimentation |
| [Galileo](https://galileo.ai/) | Agent evaluation, observability, guardrails, and quality intelligence |
| [Patronus AI](https://www.patronus.ai/) | Evaluation, red teaming, and reliability |
| [HoneyHive](https://www.honeyhive.ai/) | Evaluation, tracing, prompt management, and experiments |
| [Maxim AI](https://www.getmaxim.ai/) | Simulation, evaluation, observability, and agent quality |
| [Laminar](https://www.lmnr.ai/) | Open-source tracing, evaluation, and data collection |
| [LangWatch](https://langwatch.ai/) | Open-source LLM/agent evaluation and observability |
| [Lunary](https://lunary.ai/) | Open-source monitoring, prompts, analytics, and evaluation |
| [Portkey](https://portkey.ai/) | AI gateway plus observability, governance, and prompt management |

### Evaluation and testing frameworks

| Player | Primary fit |
|---|---|
| [DeepEval](https://deepeval.com/) | Pytest-style evaluation for LLMs, RAG, and agents |
| [Ragas](https://docs.ragas.io/) | RAG and agent evaluation |
| [promptfoo](https://www.promptfoo.dev/) | Open-source testing, comparison, red teaming, and CI |
| [Giskard](https://www.giskard.ai/) | Open-source LLM testing, evaluation, and security scanning |
| [OpenAI Evals](https://github.com/openai/evals) | Open-source evaluation framework and registry |
| [Inspect AI](https://inspect.aisi.org.uk/) | UK AI Security Institute's open-source evaluation framework |
| [Evidently](https://www.evidentlyai.com/) | Open-source AI quality, testing, and monitoring |
| [Fiddler](https://www.fiddler.ai/) | Model and GenAI observability and governance |
| [Arthur](https://www.arthur.ai/) | AI performance, monitoring, evaluation, and governance |

## Layer 10: Security, identity, guardrails, and governance

Agent security includes prompt injection, data leakage, unsafe output, excessive permissions, tool misuse, insecure MCP servers, identity delegation, policy enforcement, and auditability.

### Runtime guardrails and open frameworks

| Player | Primary fit |
|---|---|
| [NVIDIA NeMo Guardrails](https://docs.nvidia.com/nemo/guardrails/) | Open-source programmable input, output, retrieval, execution, and dialog rails |
| [Guardrails AI](https://www.guardrailsai.com/) | Structured validation, validators, and runtime guardrails |
| [LLM Guard](https://protectai.com/llm-guard) | Open and commercial prompt/output scanning and sanitization |
| [Llama Guard](https://www.llama.com/docs/model-cards-and-prompt-formats/llama-guard-4/) | Meta safety-classification models |
| [OpenAI Moderation](https://platform.openai.com/docs/guides/moderation) | OpenAI content-safety classification API |
| [AWS Guardrails for Bedrock](https://aws.amazon.com/bedrock/guardrails/) | Managed content, topic, grounding, and policy controls |
| [Azure AI Content Safety](https://azure.microsoft.com/products/ai-services/ai-content-safety) | Managed content safety and prompt shields |
| [Google Model Armor](https://cloud.google.com/security/products/model-armor) | Managed prompt/response security and policy enforcement |

### Agent and GenAI security platforms

| Player | Primary fit |
|---|---|
| [Check Point Lakera](https://www.lakera.ai/) | Prompt-injection defense, runtime guardrails, and red teaming |
| [Protect AI](https://protectai.com/) | AI supply-chain security, model scanning, red teaming, and LLM protection |
| [Prompt Security](https://prompt.security/) | Enterprise GenAI and agent security |
| [Snyk / Invariant Labs](https://invariantlabs.ai/) | Agentic AI and MCP security guardrails |
| [HiddenLayer](https://hiddenlayer.com/) | Model and AI runtime threat detection |
| [Cisco AI Defense](https://www.cisco.com/site/us/en/products/security/ai-defense/) | Enterprise AI discovery, validation, runtime protection, and governance |
| [Palo Alto Prisma AIRS](https://www.paloaltonetworks.com/prisma/prisma-ai-runtime-security) | AI runtime security, posture management, and red teaming |
| [Noma Security](https://noma.security/) | AI and agent security posture, red teaming, and runtime protection |
| [Zenity](https://zenity.io/) | Security and governance for enterprise agents and copilots |
| [Pillar Security](https://www.pillar.security/) | AI application discovery, testing, and runtime protection |
| [Mindgard](https://mindgard.ai/) | AI red teaming and security testing |
| [CalypsoAI](https://calypsoai.com/) | AI security, red teaming, and runtime controls |
| [WitnessAI](https://witness.ai/) | Enterprise AI visibility, policy, and security |
| [Credo AI](https://www.credo.ai/) | AI governance, risk, policy, and compliance |
| [Holistic AI](https://www.holisticai.com/) | AI governance, risk management, and assurance |

### Identity and authorization for agents

| Player | Primary fit |
|---|---|
| [Auth0 for AI Agents](https://auth0.com/ai) | User identity, token vaulting, and delegated authorization for agents |
| [Stytch Connected Apps](https://stytch.com/docs/guides/connected-apps/overview) | OAuth and machine/application authorization |
| [WorkOS](https://workos.com/) | Enterprise identity and authorization infrastructure |
| [Permit.io](https://www.permit.io/) | Fine-grained authorization and policy for applications and agents |
| [Oso](https://www.osohq.com/) | Authorization as a service and policy engine |
| [Cerbos](https://www.cerbos.dev/) | Open-source authorization and policy decision points |
| [Open Policy Agent](https://www.openpolicyagent.org/) | General-purpose policy engine |

## Layer 11: Model gateways, routers, and inference control planes

Gateways normalize providers and add routing, fallbacks, caching, budgets, keys, policy, logging, and usage control. They do not replace an agent runtime.

### AI-native gateways and model routers

| Player | Primary fit |
|---|---|
| [OpenRouter](https://openrouter.ai/) | Multi-provider model marketplace and routing API |
| [Vercel AI Gateway](https://vercel.com/ai-gateway) | Unified model access integrated with Vercel AI SDK and platform |
| [Cloudflare AI Gateway](https://developers.cloudflare.com/ai-gateway/) | Edge gateway, caching, logging, resilience, routing, and policy |
| [LiteLLM](https://www.litellm.ai/) | Open-source OpenAI-compatible proxy and provider normalization |
| [Portkey AI Gateway](https://portkey.ai/features/ai-gateway) | Gateway, routing, governance, observability, and prompt operations |
| [Helicone AI Gateway](https://www.helicone.ai/ai-gateway) | Open gateway with routing and observability |
| [Unify](https://unify.ai/) | Performance/cost-aware model routing and unified API |
| [Requesty](https://www.requesty.ai/) | Multi-provider routing, fallbacks, and observability |
| [Martian](https://withmartian.com/) | Model routing and optimization |
| [Not Diamond](https://www.notdiamond.ai/) | Model selection and routing |
| [Eden AI](https://www.edenai.co/) | Unified API across AI providers and capabilities |
| [TrueFoundry AI Gateway](https://www.truefoundry.com/ai-gateway) | Enterprise gateway, governance, routing, and self-hosting |

### API gateways with AI capabilities

| Player | Primary fit |
|---|---|
| [Kong AI Gateway](https://konghq.com/products/kong-ai-gateway) | Enterprise API gateway extended for LLM traffic |
| [Gloo AI Gateway](https://www.solo.io/products/gloo-ai-gateway) | Envoy-based enterprise AI gateway |
| [Tyk AI Studio / Gateway](https://tyk.io/tyk-ai-studio/) | API management and AI gateway controls |
| [Gravitee AI Gateway](https://www.gravitee.io/platform/ai-gateway) | API management, AI governance, and traffic control |
| [Zuplo](https://zuplo.com/) | Developer API gateway with AI and MCP features |
| [WSO2 AI Gateway](https://wso2.com/api-manager/ai-gateway/) | Enterprise API and AI gateway |
| [Apigee](https://cloud.google.com/apigee) | Google Cloud enterprise API management with AI integrations |
| [Envoy AI Gateway](https://aigateway.envoyproxy.io/) | Open-source Kubernetes/Envoy AI gateway |

## Layer 12: Model providers, inference clouds, and local runtimes

This is where model inference occurs. Some providers sell proprietary models, some host open models, and some supply inference software for your own compute.

### Foundation-model providers and model platforms

| Player | Primary offering |
|---|---|
| [OpenAI](https://openai.com/api/) | GPT, reasoning, realtime, image, audio, embeddings, and agent APIs |
| [Anthropic](https://www.anthropic.com/api) | Claude models and agent/coding ecosystem |
| [Google Gemini](https://ai.google.dev/) | Gemini models, multimodal APIs, and Google AI Studio |
| [Amazon Bedrock](https://aws.amazon.com/bedrock/) | Managed access to Amazon and third-party foundation models |
| [Microsoft Foundry Models](https://azure.microsoft.com/products/ai-foundry/models) | Azure model catalog and managed inference |
| [Mistral AI](https://mistral.ai/) | Frontier and open-weight models and APIs |
| [xAI](https://x.ai/api) | Grok models and APIs |
| [Cohere](https://cohere.com/) | Enterprise language models, embeddings, and reranking |
| [DeepSeek](https://www.deepseek.com/) | Reasoning, coding, and general models |
| [Qwen](https://qwen.ai/) | Alibaba's open and hosted multimodal/model family |
| [AI21 Labs](https://www.ai21.com/) | Enterprise language models and orchestration |
| [IBM watsonx.ai](https://www.ibm.com/products/watsonx-ai) | Enterprise model platform and Granite models |
| [NVIDIA NIM](https://www.nvidia.com/en-us/ai/) | Optimized model microservices and enterprise inference |
| [Hugging Face](https://huggingface.co/) | Model hub, datasets, endpoints, and inference services |

### Hosted inference for open and custom models

| Player | Primary fit |
|---|---|
| [Together AI](https://www.together.ai/) | Hosted open-model inference and fine-tuning |
| [Fireworks AI](https://fireworks.ai/) | Fast hosted inference, fine-tuning, and compound AI |
| [Groq](https://groq.com/) | Low-latency inference on LPU hardware |
| [Cerebras Inference](https://www.cerebras.ai/inference) | High-speed model inference |
| [SambaNova Cloud](https://sambanova.ai/products/sambacloud) | Enterprise model inference and platform |
| [Replicate](https://replicate.com/) | Hosted model execution and deployment |
| [Baseten](https://www.baseten.co/) | Production model inference and deployment |
| [Modal](https://modal.com/) | Serverless model and GPU workloads |
| [RunPod](https://www.runpod.io/) | GPU cloud, serverless inference, and endpoints |
| [CoreWeave](https://www.coreweave.com/) | GPU cloud and AI infrastructure |
| [Lambda](https://lambda.ai/) | GPU cloud and training/inference infrastructure |

### Self-hosted inference runtimes

| Player | Primary fit |
|---|---|
| [Ollama](https://ollama.com/) | Simple local model runtime and API |
| [vLLM](https://vllm.ai/) | High-throughput open-source serving engine |
| [SGLang](https://docs.sglang.ai/) | High-performance serving and structured generation runtime |
| [Hugging Face Text Generation Inference](https://huggingface.co/docs/text-generation-inference/) | Production serving for Hugging Face models |
| [llama.cpp](https://github.com/ggml-org/llama.cpp) | Efficient local inference across CPUs and GPUs |
| [LM Studio](https://lmstudio.ai/) | Desktop local-model discovery, execution, and API server |
| [LocalAI](https://localai.io/) | OpenAI-compatible local inference stack |
| [NVIDIA TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM) | NVIDIA-optimized LLM inference |
| [KServe](https://kserve.github.io/website/) | Kubernetes-native model serving |
| [BentoML](https://www.bentoml.com/) | Model packaging, serving, and deployment |

## Layer 13: Compute, storage, network, and deployment substrate

Every upper layer eventually runs on conventional infrastructure. The agent ecosystem does not remove this layer; it adds more dynamic workloads, secrets, egress, state, and isolation requirements.

| Category | Representative players |
|---|---|
| Hyperscale cloud | [AWS](https://aws.amazon.com/), [Microsoft Azure](https://azure.microsoft.com/), [Google Cloud](https://cloud.google.com/), [Oracle Cloud](https://www.oracle.com/cloud/) |
| Edge/serverless | [Cloudflare](https://www.cloudflare.com/developer-platform/), [Vercel](https://vercel.com/), [Fastly](https://www.fastly.com/), [Netlify](https://www.netlify.com/) |
| Application platforms | [Fly.io](https://fly.io/), [Railway](https://railway.com/), [Render](https://render.com/), [Heroku](https://www.heroku.com/), [Northflank](https://northflank.com/) |
| Containers and orchestration | [Docker](https://www.docker.com/), [Kubernetes](https://kubernetes.io/), [Nomad](https://developer.hashicorp.com/nomad), [OpenShift](https://www.redhat.com/en/technologies/cloud-computing/openshift) |
| Databases and object storage | [PostgreSQL](https://www.postgresql.org/), [Supabase](https://supabase.com/), [Neon](https://neon.com/), [MongoDB](https://www.mongodb.com/), [S3](https://aws.amazon.com/s3/), [Cloudflare R2](https://developers.cloudflare.com/r2/) |
| Messaging and streaming | [Kafka](https://kafka.apache.org/), [Redpanda](https://www.redpanda.com/), [NATS](https://nats.io/), [RabbitMQ](https://www.rabbitmq.com/), [Confluent](https://www.confluent.io/) |
| Secrets and key management | [HashiCorp Vault](https://www.vaultproject.io/), [AWS Secrets Manager](https://aws.amazon.com/secrets-manager/), [Azure Key Vault](https://azure.microsoft.com/products/key-vault), [Google Secret Manager](https://cloud.google.com/security/products/secret-manager), [Infisical](https://infisical.com/) |

## Where Goose and OpenCode sit

Goose and OpenCode are best understood as **agent products/hosts**, not merely SDKs:

```text
User
  |
  +-- Goose desktop / CLI / API
  |     +-- agent loop
  |     +-- model-provider adapters
  |     +-- MCP extensions
  |     +-- built-in code and shell tools
  |     +-- recipes, scheduling, subagents, and ACP delegation
  |
  +-- OpenCode terminal / desktop / IDE
        +-- client/server agent architecture
        +-- model-provider adapters
        +-- LSP-aware coding tools
        +-- permissions and custom agents
        +-- MCP and SDK integration
```

They compete most directly with Claude Code, Codex, Gemini CLI, Cline, Roo Code, Kilo Code, Aider, Continue, and OpenHands. They do **not** directly replace an agent deployment platform such as AgentCore, a sandbox such as Daytona, a frontend framework such as CopilotKit, or a gateway such as LiteLLM.

Goose is broader and more automation-oriented: it is positioned as a general-purpose local agent with recipes, extensions, scheduling, and the ability to conduct other agents through ACP. OpenCode is more deliberately centered on the terminal/IDE software-development experience and a provider-neutral client/server architecture.

## Common stack combinations

### Open-source local coding agent

```text
OpenCode or Goose
  -> MCP tools
  -> OpenRouter or LiteLLM
  -> Anthropic, OpenAI, Gemini, Qwen, or local Ollama/vLLM
```

### Embedded enterprise copilot

```text
CopilotKit
  -> LangGraph, Strands, Microsoft Agent Framework, or OpenAI Agents SDK
  -> Composio, Arcade, or internal MCP servers
  -> Mem0, Zep, or a vector database
  -> AgentCore, Foundry Agent Service, or Vertex Agent Engine
  -> model gateway
  -> model provider
```

### Secure coding or data-analysis agent

```text
Product UI
  -> agent framework
  -> Daytona, E2B, Runloop, or another sandbox
  -> repository, data, tests, and preview server
  -> observability and evaluation platform
  -> gateway and models
```

### Fully self-hosted stack

```text
Open WebUI or LibreChat
  -> LangGraph, PydanticAI, Mastra, or Strands
  -> self-hosted MCP servers
  -> PostgreSQL/pgvector, Qdrant, or Weaviate
  -> Langfuse and promptfoo
  -> NeMo Guardrails and OPA
  -> LiteLLM
  -> vLLM, SGLang, Ollama, or llama.cpp
  -> Kubernetes or conventional compute
```

## Selection checklist

Do not choose one winner across the whole landscape. Choose one or two candidates per required layer:

1. Is the deliverable a finished agent product or a framework for building one?
2. Is the primary interface an IDE, terminal, web application, voice channel, or API?
3. Do you need model portability or a vendor-native experience?
4. Must workflows survive restarts, wait for humans, and run for hours or days?
5. Will the agent execute arbitrary code or operate a browser?
6. Which tools require delegated user identity rather than service credentials?
7. What memory is durable, deletable, attributable, and policy-governed?
8. How will you evaluate full trajectories, not only final text?
9. Where are prompt injection and tool authorization enforced?
10. Do you need self-hosting, air-gapping, regional control, or managed operations?
11. Which protocol boundaries should remain portable: MCP, A2A, AG-UI, or ACP?
12. What happens when the model, tool, browser, sandbox, or human approval times out?

## Practical shortlist by problem

| Problem | Start with |
|---|---|
| Open-source local coding agent | Goose, OpenCode, Cline, Roo Code, Aider, OpenHands |
| Vendor-native coding agent | Claude Code, Codex, Gemini CLI, GitHub Copilot |
| Embedded product copilot | CopilotKit, assistant-ui, Vercel AI SDK |
| TypeScript AI application | Vercel AI SDK, TanStack AI, Mastra |
| Python agent | PydanticAI, LangGraph, OpenAI Agents SDK, Strands, Google ADK |
| Explicit durable state machine | LangGraph plus LangSmith Deployment or a durable workflow platform |
| Multi-agent teams | Strands, CrewAI, Google ADK, Microsoft Agent Framework, AutoGen/AG2 |
| Managed AWS operation | Bedrock AgentCore |
| Managed Azure operation | Microsoft Foundry Agent Service |
| Managed Google Cloud operation | Vertex AI Agent Engine |
| Edge/stateful realtime agent | Cloudflare Agents |
| Secure code and computer execution | Daytona, E2B, Runloop, Modal, Vercel Sandbox |
| Browser automation | Browserbase, Steel, Hyperbrowser, Browser Use, AgentCore Browser |
| Agent tools and OAuth | Composio, Arcade, Pipedream, Nango, Klavis |
| Long-term agent memory | Mem0, Zep, Letta, Cognee, Graphiti |
| RAG and document context | LlamaIndex, Haystack, Unstructured, Qdrant/Weaviate/Pinecone |
| Open-source observability | Langfuse, Phoenix, Opik, MLflow, Laminar |
| Evaluation and red teaming | Braintrust, DeepEval, Ragas, promptfoo, Giskard, Inspect AI |
| Runtime security | Lakera, NeMo Guardrails, LLM Guard, Protect AI, Invariant/Snyk |
| Self-hosted model gateway | LiteLLM, Portkey, Helicone, Envoy AI Gateway |
| Broad model marketplace | OpenRouter |
| Local model inference | Ollama, vLLM, SGLang, llama.cpp, LM Studio |

## Maintenance notes

- This market changes rapidly. Recheck status, licensing, pricing, and product names before procurement.
- Open-source client libraries and self-hosted servers from the same company can use different licenses.
- A protocol implementation claim does not guarantee full interoperability; test the exact client/server pair.
- “Memory” may mean message history, semantic facts, a knowledge graph, checkpoint state, or durable workflow state. These are different products and data-governance obligations.
- “Agent platform” is heavily overloaded. Always identify whether it supplies authoring, hosting, execution, tools, UI, models, or only a control plane.
- The earlier detailed comparison remains available in [Vercel AI SDK vs TanStack AI and adjacent platforms](vercel-ai-sdk-vs-tanstack-ai.md).