# Awesome Agent Infrastructure

> A curated list of AI agent infrastructure: memory stores, vector databases, execution sandboxes, MCP servers, tool registries, observability, and evaluation tools for building and running autonomous agents.

- **Type:** MCP server
- **Install:** `agentstack add mcp-backblaze-labs-awesome-agent-infrastructure`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [backblaze-labs](https://agentstack.voostack.com/s/backblaze-labs)
- **Installs:** 0
- **Category:** [Integrations](https://agentstack.voostack.com/c/integrations)
- **Latest version:** 0.1.0
- **License:** CC0-1.0
- **Upstream author:** [backblaze-labs](https://github.com/backblaze-labs)
- **Source:** https://github.com/backblaze-labs/awesome-agent-infrastructure
- **Website:** https://blze.ai/storage

## Install

```sh
agentstack add mcp-backblaze-labs-awesome-agent-infrastructure
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Awesome Agent Infrastructure [](https://awesome.re) [](http://makeapullrequest.com) [](https://creativecommons.org/publicdomain/zero/1.0/)

A curated list of infrastructure for building reliable LLM agents — frameworks, memory, tool protocols, sandboxes, browsers, observability, and retrieval.

Maintained by [Backblaze](https://www.backblaze.com).

### Related Lists

- [Awesome ML Data Pipelines](https://github.com/backblaze-labs/awesome-ml-data-pipelines)
- [Awesome Multimodal Data](https://github.com/backblaze-labs/awesome-multimodal-data)
- [Awesome Physical AI](https://github.com/backblaze-labs/awesome-physical-ai)
- [Awesome Image Generation](https://github.com/backblaze-labs/awesome-image-generation)
- [Awesome Video Generation](https://github.com/backblaze-labs/awesome-video-generation)
- [Awesome Audio Generation](https://github.com/backblaze-labs/awesome-audio-generation)

## Contents

- [Agent Frameworks](#agent-frameworks)
- [Memory and State](#memory-and-state)
- [Tool Protocols and Servers](#tool-protocols-and-servers)
- [Execution Sandboxes](#execution-sandboxes)
- [Browser and Computer Use](#browser-and-computer-use)
- [Observability and Evaluation](#observability-and-evaluation)
- [Retrieval and RAG](#retrieval-and-rag)
- [Vector Databases](#vector-databases)
- [Templates and Example Projects](#templates-and-example-projects)

---

## Agent Frameworks

> Libraries for building LLM agents — planning, tool use, multi-agent orchestration.

- **[Microsoft AutoGen](https://microsoft.github.io/autogen/)** – Multi-agent conversation framework from Microsoft Research. AutoGen 0.4 rewrote it around an event-driven runtime. [Docs](https://microsoft.github.io/autogen/stable/) | SDK: Python (pip install autogen-agentchat)
- **[CrewAI](https://www.crewai.com)** – Role-based multi-agent framework. Agents, tasks, and tools composed into crews with deterministic or planning-based flows. [Docs](https://docs.crewai.com) | SDK: Python (pip install crewai)
- **[Agno](https://www.agno.com)** – Lightweight Python framework for building multimodal agents and agentic systems. Formerly Phidata. [Docs](https://docs.agno.com) | SDK: Python (pip install agno)
- **[LangGraph](https://langchain-ai.github.io/langgraph/)** – Graph-based agent runtime from the LangChain team. Durable execution, human-in-the-loop, and multi-actor patterns. [Docs](https://langchain-ai.github.io/langgraph/tutorials/introduction/) | SDK: Python (pip install langgraph), JS (npm install @langchain/langgraph)
- **[HuggingFace smolagents](https://huggingface.co/docs/smolagents)** – Minimal "code agent" library — agents write Python to solve tasks. ~1k LoC core; easy to audit and extend. [Docs](https://github.com/huggingface/smolagents) | SDK: Python (pip install smolagents)
- **[Mastra](https://mastra.ai)** – TypeScript-first agent framework with workflows, RAG, and evals. From the creators of Gatsby. [Docs](https://mastra.ai/docs) | SDK: TypeScript (npm install @mastra/core)
- **[OpenAI Agents SDK](https://openai.github.io/openai-agents-python/)** – Official OpenAI agent framework. Handoffs, guardrails, built-in tracing, and Responses-API-native execution. [Docs](https://github.com/openai/openai-agents-python) | SDK: Python (pip install openai-agents)
- **[Pydantic AI](https://ai.pydantic.dev)** – Agent framework from the Pydantic team. Type-safe tool calling, structured outputs, dependency injection. [Docs](https://ai.pydantic.dev) | SDK: Python (pip install pydantic-ai)
- **[AG2](https://ag2.ai)** – Community-maintained fork of AutoGen 0.2. Multi-agent conversation framework with swarms, group chats, and nested chat patterns. [Docs](https://docs.ag2.ai/) | SDK: Python (pip install ag2)
- **[AgentScope](https://agentscope.io)** – Python agent framework with an event-driven runtime, human-in-the-loop, sandboxed tool execution, and Agent-as-a-Service REST deployment. v2.0 released May 2026. [Docs](https://docs.agentscope.io) | SDK: Python (pip install agentscope)
- **[DeerFlow](https://github.com/bytedance/deer-flow)** – ByteDance's open-source super-agent harness built on LangGraph. Orchestrates sub-agents, memory, sandboxes, and skills for long-horizon tasks. [Docs](https://github.com/bytedance/deer-flow/blob/main/README.md)
- **[Flowise](https://flowiseai.com)** – Open-source visual builder for LLM agents and workflows. Drag-and-drop Agentflow canvas plus REST API, JS/Python SDK, and CLI for programmatic integration into production applications. [Docs](https://docs.flowiseai.com) | SDK: TypeScript (npm install -g flowise)
- **[Google ADK](https://adk.dev)** – Google's open-source agent development kit. Build, evaluate, and deploy multi-agent systems; multi-language with Gemini-optimized but model-agnostic. [Docs](https://adk.dev) | SDK: Python (pip install google-adk), TypeScript (npm install @google/adk)
- **[Langflow](https://www.langflow.org)** – Low-code builder for AI agents and RAG applications. Visual canvas with Python escape hatches, deploys flows as REST APIs or MCP servers; 40+ model and vector-store integrations. [Docs](https://docs.langflow.org) | SDK: Python (pip install langflow)
- **[Langroid](https://langroid.github.io/langroid/)** – Lightweight Python multi-agent framework from CMU/UW-Madison. Task-delegation via message passing; no LangChain dependency. [Docs](https://langroid.github.io/langroid/) | SDK: Python (pip install langroid)
- **[MetaGPT](https://github.com/FoundationAgents/MetaGPT)** – Multi-agent framework that assigns software-company roles (PM, architect, engineer) to LLMs. Input a requirement, get PRD, design, code, and tests. [Docs](https://docs.deepwisdom.ai/main/en/) | SDK: Python (pip install metagpt)
- **[Microsoft Agent Framework](https://devblogs.microsoft.com/agent-framework/)** – Microsoft's production-ready open-source agent SDK and runtime for Python and .NET. Unifies AutoGen orchestration and Semantic Kernel foundations. [Docs](https://learn.microsoft.com/en-us/agent-framework/overview/) | SDK: Python (pip install agent-framework), .NET (dotnet add package Microsoft.Agents.AI)
- **[open-multi-agent](https://github.com/JackChen-me/open-multi-agent)** – TypeScript multi-agent orchestration with automatic goal-to-DAG decomposition, parallel task execution, MCP integration, and live tracing. Three runtime dependencies; 10+ LLM providers supported. SDK: TypeScript (npm install @jackchen_me/open-multi-agent)
- **[OpenAI Agents SDK (TypeScript)](https://openai.github.io/openai-agents-js/)** – Official OpenAI agent framework for TypeScript and JavaScript. Agents, handoffs, guardrails, voice via Realtime API, and built-in tracing. [Docs](https://openai.github.io/openai-agents-js/) | SDK: TypeScript (npm install @openai/agents)
- **[OpenSRE](https://github.com/Tracer-Cloud/opensre)** – Open-source toolkit for building AI SRE agents. Connects to 60+ observability, cloud, and incident-management tools; auto-fetches alert context, correlates logs/metrics, and generates root-cause reports.
- **[Semantic Kernel](https://learn.microsoft.com/en-us/semantic-kernel/overview/)** – Microsoft's open-source SDK for building LLM agents and multi-agent systems. Model-agnostic; plugins, planners, and process orchestration across Python, C#, and Java. [Docs](https://learn.microsoft.com/en-us/semantic-kernel/) | SDK: Python (pip install semantic-kernel), C# (dotnet add package Microsoft.SemanticKernel), Java (Maven: com.microsoft.semantic-kernel)
- **[Strands Agents](https://strandsagents.com)** – AWS-backed open-source agent SDK. Define tools as functions; the model-driven loop handles planning and execution with no workflow graphs required. [Docs](https://strandsagents.com/latest/) | SDK: Python (pip install strands-agents), TypeScript (npm install @strands-agents/sdk)
- **[VoltAgent](https://voltagent.dev)** – TypeScript agent framework with memory adapters, RAG, tool registry, multi-agent supervisor coordination, voice support, and built-in evals. [Docs](https://voltagent.dev/docs/) | SDK: TypeScript (npm create voltagent-app@latest)

## Memory and State

> Long-term memory, session state, and knowledge-retention layers for agents.

- **[Mem0](https://mem0.ai)** – Memory layer for AI agents. Personalization through user/agent/session memories with semantic recall. [Docs](https://docs.mem0.ai) | SDK: Python (pip install mem0ai), Node (npm install mem0ai)
- **[Letta](https://www.letta.com)** – Open-source agent server focused on long-term memory. Successor to MemGPT; agents are first-class stateful services. [Docs](https://docs.letta.com) | SDK: Python (pip install letta-client)
- **[Zep](https://www.getzep.com)** – Memory and context platform for LLM apps. Knowledge-graph-backed user memory with temporal reasoning. [Docs](https://help.getzep.com)
- **[Cognee](https://www.cognee.ai)** – Knowledge engine for agent memory. ECL pipeline ingests any data into a hybrid vector + knowledge graph for structured, traceable recall. [Docs](https://docs.cognee.ai) | SDK: Python (pip install cognee)
- **[Graphiti](https://github.com/getzep/graphiti)** – Open-source temporal context graph engine. Tracks how facts change over time with full provenance; hybrid semantic + keyword + graph retrieval. [Docs](https://help.getzep.com/graphiti) | SDK: Python (pip install graphiti-core)
- **[Hindsight](https://hindsight.vectorize.io)** – Open-source agent memory system using biomimetic data structures. Organises memories into world facts, experiences, and mental models; TEMPR retrieval combines semantic, keyword, graph, and temporal search. [Docs](https://hindsight.vectorize.io/docs) | SDK: Python (pip install hindsight-client), TypeScript (npm install @vectorize-io/hindsight-client)
- **[Honcho](https://honcho.dev)** – Memory infrastructure for stateful agents. Stores messages to per-peer sessions, runs background reasoning to build user representations, and returns curated context via a fast query API. [Docs](https://docs.honcho.dev) | SDK: Python (uv add honcho-ai)
- **[Puppyone](https://www.puppyone.ai)** – File system for agents. Connect, govern, version, and share context across agent workflows. [Docs](https://www.puppyone.ai/doc)
- **[Redis Agent Memory Server](https://github.com/redis/agent-memory-server)** – Memory layer for AI agents backed by Redis. Two-tier working + long-term memory, semantic/keyword/hybrid search, REST and MCP interfaces, and multi-LLM provider support. [Docs](https://github.com/redis/agent-memory-server/blob/main/README.md) | SDK: Python (pip install agent-memory-client)
- **[ReMe](https://github.com/agentscope-ai/ReMe)** – Memory management kit for AI agents. Conversation compaction, long-term file-based and vector memory, semantic search; compresses context by up to 99.5% while retaining critical facts. [Docs](https://github.com/agentscope-ai/ReMe/blob/main/README.md) | SDK: Python (pip install reme)
- **[Supermemory](https://supermemory.ai)** – Memory and context API for AI agents. Ingests documents and conversations, extracts facts, builds user profiles, and returns relevant context via hybrid semantic search; SDK and REST interfaces. [Docs](https://docs.supermemory.ai) | SDK: TypeScript (npm install @supermemory/sdk), Python

## Tool Protocols and Servers

> Standardised interfaces for exposing tools and data sources to agents (MCP and friends).

- **[MCP Reference Servers](https://github.com/modelcontextprotocol/servers)** – Reference MCP server implementations for filesystem, Git, GitHub, SQL, Slack, and more.
- **[MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk)** – Official Python SDK for building and consuming MCP servers and clients. SDK: Python (pip install mcp)
- **[MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)** – Official TypeScript SDK for MCP servers and clients. SDK: TypeScript (npm install @modelcontextprotocol/sdk)
- **[Anthropic Model Context Protocol](https://modelcontextprotocol.io)** – Open protocol for connecting AI applications to tools and data sources. Spec, reference servers, and official SDKs. [Docs](https://modelcontextprotocol.io/introduction)
- **[Agent2Agent Protocol (A2A)](https://github.com/a2aproject/A2A)** – Open protocol for communication and interoperability between AI agents. JSON-RPC 2.0 over HTTP with SDKs for Python, Go, JS, Java, and .NET. [Docs](https://a2a-protocol.org)
- **[AWS MCP Servers](https://awslabs.github.io/mcp/)** – Suite of 53 open-source MCP servers for AWS services — CloudFormation, Bedrock, DynamoDB, EKS, S3, and more. [Docs](https://awslabs.github.io/mcp/)
- **[Composio](https://composio.dev)** – Tool-integration SDK for AI agents. 1000+ pre-built tool connectors (GitHub, Slack, Jira, etc.) with managed auth and sandboxed execution. [Docs](https://docs.composio.dev) | SDK: Python (pip install composio), TypeScript (npm install @composio/core)
- **[Headroom](https://github.com/chopratejas/headroom)** – Context compression layer for AI agents. Compresses tool outputs, logs, RAG chunks, and files 60-95% before they reach the LLM; runs as a library, proxy, or MCP server with reversible compression. SDK: Python (pip install headroom)
- **[IBM ContextForge](https://ibm.github.io/mcp-context-forge/latest/)** – Open-source MCP/A2A/REST gateway and registry. Federates MCP servers, A2A agents, and REST/gRPC APIs behind a single governed endpoint with auth, rate limiting, and OpenTelemetry tracing. [Docs](https://ibm.github.io/mcp-context-forge/latest/) | SDK: Python (pip install mcp-contextforge-gateway)
- **[MCP Inspector](https://github.com/modelcontextprotocol/inspector)** – Interactive visual tool for testing and debugging MCP servers. Supports STDIO, SSE, and Streamable HTTP transports. SDK: TypeScript (npx @modelcontextprotocol/inspector)
- **[MCPX](https://github.com/TheLunarCompany/lunar/tree/main/mcpx)** – Open-source MCP gateway and aggregator. Consolidates multiple MCP servers behind a single governed entry point with rate limiting and traffic policies. [Docs](https://docs.lunar.dev/mcpx/)
- **[n8n-mcp](https://github.com/czlonkowski/n8n-mcp)** – MCP server exposing n8n's 1,650+ workflow nodes to AI agents. Provides node docs, schema properties, operations, and workflow validation for agents building n8n automations. SDK: TypeScript (npx n8n-mcp)

## Execution Sandboxes

> Secure environments for running agent-generated code, shell commands, and browser sessions.

- **[Daytona](https://www.daytona.io)** – Open-source dev-environment manager; Daytona Sandboxes expose a sandbox API for agents and CI pipelines. [Docs](https://www.daytona.io/docs)
- **[E2B](https://e2b.dev)** – Secure cloud sandboxes for running AI-generated code. Firecracker microVMs, sub-second startup, per-session isolation. [Docs](https://e2b.dev/docs) | SDK: Python (pip install e2b), JS (npm install @e2b/code-interpreter)
- **[Microsoft Agent Governance Toolkit](https://github.com/microsoft/agent-governance-toolkit)** – Runtime policy enforcement for autonomous agents. Zero-trust identity, execution sandboxing, sub-millisecond policy checks; covers all 10 OWASP Agentic Top 10 risks. SDK: Python (pip install agent-governance-toolkit), TypeScript (npm install @microsoft/agentmesh-sdk)
- **[Modal Sandboxes](https://modal.com/docs/guide/sandbox)** – Serverless sandbox primitive inside Modal. Arbitrary container execution, ephemeral filesystems, strict network policies.
- **[OpenShell](https://github.com/NVIDIA/OpenShell)** – NVIDIA's open-source sandbox runtime for autonomous agents. Declarative YAML policies govern file access, network activity, and data exfiltration; supports Claude, Codex, Copilot, and OpenCode. SDK: Python (uv tool install openshell)
- **[Riza](https://riza.io)** – Secure code-execution API for LLM tool calls. Python, JS, PHP, Ruby; strict WASM-based isolation. [Docs](https://docs.riza.io)

## Browser and Computer Use

> Platforms and SDKs that let agents drive web browsers and full desktops.

- **[browser

…

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [backblaze-labs](https://github.com/backblaze-labs)
- **Source:** [backblaze-labs/awesome-agent-infrastructure](https://github.com/backblaze-labs/awesome-agent-infrastructure)
- **License:** CC0-1.0
- **Homepage:** https://blze.ai/storage

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-backblaze-labs-awesome-agent-infrastructure
- Seller: https://agentstack.voostack.com/s/backblaze-labs
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.