InfraA field guide for builders
Infrastructure
for AI agents.
Find the tools behind capable agents. Sandboxes, browsers, memory, and more. Hosted or in your own stack.
40 tools · 6 categories · A growing little ecosystem
Start with a capability
Every agent needs a few good tools.
Sandboxes and execution
Run agent-generated code in isolated environments.
6 toolsBrowsers and computer use
Give agents browsers they can navigate and control.
9 toolsSearch and web data
Find information on the web and turn pages into usable data.
5 toolsMemory and state
Save and retrieve context across agent sessions.
7 toolsTools and authentication
Connect agents to applications and control their access.
6 toolsObservability and evaluation
Inspect agent behavior and evaluate application quality.
The directory
Find your next building block
Listed A–Z. Hosted and self-hosted options share one listing. About the labels
AgentOps
AgentOpsAgent observability software for recording runs, inspecting execution graphs, and monitoring application activity and model usage.
Details & sources
The repository includes a self-hosted dashboard and API backend, alongside the SDK.
Source repository ↗ · AgentOps repository including the app and backend.
Reviewed
Anchor Browser
Anchor BrowserCloud browser infrastructure for agents, with programmatic session control and browser authentication workflows.
HostedDetails & sources
Reviewed
Apify
ApifyA managed platform for running reusable data-collection tools called Actors, including website crawling and structured extraction.
HostedDetails & sources
Individual Actors have their own scope and terms; open-source crawler libraries do not make the Apify platform self-hostable.
Reviewed
Arcade
ArcadeAn agent action platform for connecting application tools and enforcing authentication, access policies, and governance around tool execution.
HostedSelf-hostedDetails & sources
Self-hosting is documented through a Helm deployment; that does not imply an open-source platform license.
Reviewed
Arize Phoenix
Arize AISelf-hosted observability and evaluation software for tracing agent applications and running experiments with datasets.
Self-hostedSource availableElastic-2.0Details & sources
The reviewed Phoenix runtime uses Elastic License 2.0. Arize AX is a separate managed product.
Source repository ↗ · Phoenix application runtime; some related packages may have different licenses.
Reviewed
Auth0 for AI Agents
Auth0Identity capabilities for AI agents, including registered agent principals and delegated access tracked through tokens and audit records.
HostedDetails & sources
Reviewed
Braintrust
BraintrustAn agent observability platform for collecting traces, exploring production behavior, and evaluating changes against datasets.
HostedSelf-hosted data planeDetails & sources
Enterprise self-hosted data plane. The UI, authentication, and control plane remain managed by Braintrust.
Reviewed
Brave Search API
BraveSearch APIs backed by Brave's independent index, including web results and context-oriented responses for AI applications.
HostedDetails & sources
Reviewed
Bright Data Web Scraper APIs
Bright DataManaged scraper APIs that retrieve structured data from supported websites for applications and data workflows.
HostedDetails & sources
Reviewed
Browser Use Browser Infrastructure
Browser UseManaged browser sessions that an application or agent can launch and control through its own code over CDP.
HostedDetails & sources
The managed browser product is distinct from Browser Use Agents and the open-source local agent library.
Reviewed
Browserbase
BrowserbaseManaged browser infrastructure for creating, controlling, and observing browser sessions used by AI agents.
HostedDetails & sources
This listing covers browser infrastructure, rather than the provider's model access or other platform products.
Reviewed
Browserless
BrowserlessBrowser infrastructure for running automation with tools such as Playwright and Puppeteer, available as a managed service or a Docker deployment.
Details & sources
The runtime uses SSPL or a commercial license. Self-hosted and managed editions have different feature and licensing terms.
Source repository ↗ · Browserless runtime under its dual-license terms.
Reviewed
Cerbos
CerbosA policy-based authorization engine for deciding what users, services, and AI agents can do with application resources.
Self-hostedOpen sourceApache-2.0Details & sources
This listing covers the self-hosted Policy Decision Point. Cerbos Hub provides a separate policy-management service; authentication is still handled by an identity provider.
Source repository ↗ · Cerbos Policy Decision Point; Cerbos Hub has separate terms.
Reviewed
Cloudflare Sandboxes
CloudflareIsolated environments accessed from Cloudflare Workers for executing generated code and workloads that should run separately from the application.
HostedDetails & sources
Requires a Cloudflare Workers setup. The sandbox SDK source does not establish that the Cloudflare service can be self-hosted.
Reviewed
Cognee
CogneeAgent memory software that turns documents and conversations into persistent knowledge graphs that agents can search and reuse.
Details & sources
Local operation and a managed Cognee Cloud offering are documented; deployment choices and model requirements vary by configuration.
Source repository ↗ · Cognee core memory platform. Managed service terms are separate.
Reviewed
Composio
ComposioAn integration platform that gives agents application tools with per-user connections, managed authentication, and tool discovery.
HostedDetails & sources
The public Composio repository is an SDK monorepo; its MIT license does not establish a self-hostable platform backend.
Reviewed
Daytona Sandboxes
DaytonaManaged sandbox computers for running agent-generated code, with programmatic process execution, file access, and persistent snapshots.
HostedDetails & sources
This listing covers the current sandbox service; the archived daytonaio/daytona repository is not evidence of a current self-hosted runtime.
Reviewed
DeepEval
Confident AIAn evaluation framework for running tests against agent outputs, trajectories, tool use, and individual application components.
Self-hostedOpen sourceApache-2.0Details & sources
The framework runs in your environment. Confident AI is a separate optional hosted platform; model-based evaluators may call configured external providers.
Source repository ↗ · DeepEval evaluation framework; optional Confident AI service terms are separate.
Reviewed
Descope Agentic Identity Hub
DescopeIdentity infrastructure for agents and MCP servers, including delegated access, credential management, consent, and policy controls.
HostedDetails & sources
Reviewed
E2B
E2BSandbox infrastructure that gives agents isolated Linux environments for running code, working with files, and pausing or resuming sessions.
Details & sources
The public runtime includes the sandbox backend; running it requires operating its infrastructure.
Source repository ↗ · E2B Runtime backend. Hosted service terms are separate.
Reviewed
Exa
ExaWeb search and page-content APIs for giving agents current information from the web.
HostedDetails & sources
Reviewed
Firecrawl
FirecrawlWeb search, scraping, and crawling tools that turn pages into Markdown and structured data for AI applications.
Details & sources
Some managed-service capabilities are unavailable or require additional setup in the self-hosted edition.
Source repository ↗ · Firecrawl server code. SDKs and the hosted service have separate terms.
Reviewed
Graphiti
ZepA framework for building and querying temporal context graphs, so agents can retrieve facts and their changes over time.
Self-hostedOpen sourceApache-2.0Details & sources
Runs with a configured graph database and model/embedding providers. Zep's managed service is a separate product.
Source repository ↗ · Graphiti context-graph framework and included server components.
Reviewed
Jina Reader
Jina AIURL reading and web search interfaces that return text suitable for agent and retrieval workflows.
Details & sources
The public server supports stateless or bucket-cached deployments; the hosted service's storage layer is not included.
Source repository ↗ · Public Reader server branch, excluding the hosted MongoDB storage layer.
Reviewed
Kernel
KernelCloud browser infrastructure that agents can control through browser automation protocols or computer controls, with session viewing and replay tools.
HostedDetails & sources
Public browser-image components do not establish availability of the complete hosted platform for self-hosting.
Reviewed
Langfuse
LangfuseAgent observability and evaluation software for inspecting traces, working with datasets, and testing application behavior.
Details & sources
Community code is MIT-licensed; enterprise directories and some add-on features require separate licensing.
Source repository ↗ · Community code outside ee/, web/src/ee/, and worker/src/ee/. Enterprise code has a separate license.
Reviewed
LangSmith
LangChainAn observability and evaluation platform for tracing agent runs, monitoring production behavior, and working with evaluation datasets.
HostedSelf-hostedDetails & sources
Enterprise plan add-on requiring a license key; the platform runs in the customer infrastructure.
Reviewed
Mem0
Mem0Memory infrastructure for saving and retrieving context across agent sessions, offered as a managed platform or a self-hosted library and server.
Details & sources
Self-hosted deployments require choosing and operating the configured memory components.
Source repository ↗ · Mem0 open-source memory engine and server. Hosted platform terms are separate.
Reviewed
Microsandbox
Microsandbox projectBetaA local microVM runtime for running untrusted agent workloads and code through a CLI or an embedded SDK.
Self-hostedOpen sourceApache-2.0Details & sources
The project describes itself as beta software. It requires a supported virtualization environment; consult the platform requirements in its documentation.
Source repository ↗ · Microsandbox local microVM runtime.
Reviewed
Modal Sandboxes
ModalProgrammatically created sandbox environments on Modal for executing commands and running untrusted or generated code.
HostedDetails & sources
This listing covers Sandboxes rather than Modal inference or model-serving products.
Reviewed
Nango
NangoIntegration infrastructure for connecting products and agents to external APIs, with credential management and integration execution.
Details & sources
The public runtime uses Elastic License 2.0; self-hosted feature availability differs from the managed and commercial offerings.
Source repository ↗ · Nango runtime under Elastic License 2.0; paid features and terms remain separate.
Reviewed
OpenSandbox
AlibabaSelf-hosted sandbox infrastructure for agent code execution, command and file operations, with local Docker and Kubernetes deployment options.
Self-hostedOpen sourceApache-2.0Details & sources
Requires operating the sandbox server and its container or Kubernetes runtime.
Source repository ↗ · OpenSandbox server, runtimes, and project components.
Reviewed
Pipedream Connect
PipedreamManaged application integrations for agents, combining connected user accounts, prebuilt tools, and authenticated API access.
HostedDetails & sources
Reviewed
Runloop Devboxes
RunloopManaged sandbox development environments for software agents, with reusable environment templates and snapshot-based session persistence.
HostedDetails & sources
Reviewed
ScrapingBee
ScrapingBeeA web scraping API that handles browser rendering and proxy infrastructure and can return structured data or Markdown.
HostedDetails & sources
Reviewed
SerpApi
SerpApiSearch-results APIs that return structured results from search engines for use in applications and agent workflows.
HostedDetails & sources
Reviewed
Steel Browser
SteelBetaBrowser infrastructure with a session API for agent-driven web automation, offered through Steel Cloud or a self-hosted browser server.
Details & sources
The reviewed browser project describes itself as public beta and includes Docker and local deployment instructions.
Source repository ↗ · Steel browser API and UI server; managed cloud terms are separate.
Reviewed
Supermemory
SupermemoryMemory and context infrastructure that extracts, stores, and retrieves information for agents across sessions.
HostedSelf-hostedDetails & sources
The downloadable self-hosted server is built from non-public code. The public repository's MIT license covers SDKs and other components, not that server.
Reviewed
Tavily
TavilyWeb access APIs for agent search, content extraction, and crawling, returning information for application and research workflows.
HostedDetails & sources
Reviewed
Zep
ZepManaged context infrastructure that connects business records, documents, and conversations and retrieves relevant context for agents.
HostedDetails & sources
Graphiti is listed separately as a self-hosted temporal context-graph framework; it is not the same product as Zep's managed context infrastructure.
Reviewed
No tools match those filters.
Try a different search or see all deployment options.
A different way to browse
Have a tool in mind?
Start with a familiar name and explore other tools in its category.
Missing a building block?
Help another builder find their next useful tool.