Agent-Native Tools
Essential infrastructure for autonomous companies. Curated tools for AI agents.
NVIDIA NemoClaw
Policy-based guardrails for autonomous AI agents — the missing security layer
NemoClaw is NVIDIA's official security and privacy stack for OpenClaw agents, announced at GTC 2026. It installs in a single command and adds NVIDIA OpenShell — an open-source runtime that enforces policy-based guardrails, sandboxes AI models, and adds data privacy protections. Built alongside CrowdStrike, Cisco, and Microsoft Security. Runs on any platform: cloud, on-premises, RTX PCs, DGX Station, DGX Spark.
Perplexity Computer
19 AI models, one coordinator — multi-model orchestration as infrastructure
Perplexity Computer is a multi-model agent orchestration platform launched February 2026. It accepts a high-level objective, breaks it into subtasks, and routes each subtask to whichever AI model is best suited for it — Claude Opus 4.6 for orchestration and coding, Gemini for research and image generation, GPT-5.2 for long-context recall and web search, Grok for speed-sensitive tasks, and 16 others. The coordinator decides which specialist handles each job. A Personal Computer variant runs 24/7 on a local Mac Mini with full file and app access.
Hermes Agent
Open source agent that learns and grows with multi-level memory
An open source AI agent from Nous Research that remembers what it learns and gets more capable over time. Features a multi-level memory system and persistent dedicated machine access, allowing the agent to build long-term knowledge and maintain state across sessions.
LM Studio
Run local LLMs on your desktop with zero cloud dependencies
Desktop application for running large language models locally on Mac, Windows, and Linux. Download, configure, and chat with models like Llama, Gemma, Qwen, and DeepSeek without sending data to external servers or paying API costs.
Raindrop
Sentry for AI Agents — observability and monitoring
The first Sentry-like monitoring platform for AI agents. Detect silent failures, tool malfunctions, user frustration, and abnormal trajectories in real-time. Deep search across millions of events, A/B test prompts and models, and get Slack alerts when things go wrong.
Zo Computer
AI-powered cloud computer that works 24/7
Zo Computer is your personal AI companion running on a private cloud server. Text what you need and Zo builds it — apps, automations, internal tools, dashboards — deployed to your own Linux server without touching a config file or writing code.
Outseta
All-in-one membership stack that runs without human ops
Outseta is an all-in-one membership management platform that automates payments, authentication, content gating, CRM, email marketing, and support — replacing the entire membership operations human role with a single integrated system.
Manus
Full-stack web apps from natural language
Manus is an AI website builder that creates full-stack web applications from natural language prompts. Describe what you want, and it generates production-ready code with built-in database, backend, SEO optimization, and analytics — all without human developers.
Replit
Deploy apps from natural language without touching code
Replit is an AI-native development platform where an Agent builds full-stack apps from natural language prompts. Describe what you want, and it generates production-ready code, handles infrastructure, and deploys instantly — no local setup, no configuration, no human developers needed in the loop.
SelfClaw
Sybil-resistant verification for AI agents
SelfClaw is the trust layer for AI agent economies. Verify agents via zero-knowledge proofs linked to real humans, enable self-custody wallets for autonomous transactions, and deploy tokenized agent economies — all while maintaining privacy and preventing spam.
PicoClaw
$10 OpenClaw alternative running on microcontrollers
PicoClaw is an ultra-lightweight personal AI assistant that runs on $10 hardware using 99% less memory than OpenClaw. Written in Go through a self-bootstrapping process where the AI agent helped design its own architecture. The result is a single portable binary (under 10MB) that runs across ARM, x86, and RISC-V platforms and starts in about one second, even on slow single-core processors.
Do Anything
Autonomous AI agent that actually does the work
Do Anything is not a chatbot or assistant — it's an autonomous agent that accomplishes real work. Your agent can send emails, make phone calls with AI voice, browse websites, write code, create documents, build websites, generate images, spawn subagents, and connect to 3,000+ apps. It can even create its own identity with unique name and email to sign up for services and manage social media accounts.
Kimi Claw
OpenClaw native in your browser with agent swarm
Kimi Claw brings OpenClaw directly into the Kimi AI platform. Access 5,000+ community skills from ClawHub, get 40GB cloud storage, and leverage pro-grade search — all living in your browser tab, online 24/7. Built on Kimi K2.5, one of the strongest open-source models for visual coding.
AgentMail
Email inboxes for AI agents
AgentMail is the first email provider built specifically for AI agents. Create, manage, and operate email inboxes entirely via API — giving your agents their own email addresses for two-way conversations.
Bankr
AI agents that manage crypto, launch tokens, and fund their own compute
Bankr is the infrastructure layer for self-sustaining AI agents. Every agent gets cross-chain wallets, can launch tokens via fair launch, and earns trading fees that automatically fund compute costs — all without human financial management.
Supermemory
Universal Memory API for AI apps
Supermemory gives your agents state-of-the-art memory, RAG, user profiles, connectors, and extractors — all built in. Extremely low latency. Works with any model. Built on Postgres and Cloudflare Durable Objects, it scales to 50 million tokens per user and handles over 5 billion tokens daily.
Nebula
Your AI employee that manages your inbox, syncs tools, and automates workflows
Nebula is an AI employee platform that runs 24/7 to manage your inbox, sync your tools, and automate your workflows. Built like Slack but with intelligent agents instead of humans — each channel represents a business workflow with agents that have access to internet search, code execution, browser automation, and integrations with Google Suite, GitHub, Slack, Notion, Linear, and more.
Spacebot
AI operating system for autonomous agent teams
Multi-agent architecture with persistent memory graph. Purpose-built infrastructure for running agent swarms without human coordination.
Tito
Event management platform that agents can actually use
Tito is an event management platform designed with API-first architecture and clean email-based authentication. Unlike most event platforms that block automated access with CAPTCHAs and complex flows, Tito allows AI agents to create events, manage RSVPs, and track attendance programmatically.
Parallel
Highest accuracy web search API purpose-built for AI agents
Parallel is a web search and research API engineered specifically for AI agents. It delivers production-ready outputs built on cross-referenced facts with minimal hallucination, predictable per-query pricing, and verifiable provenance for every atomic output. Powers millions of daily requests with SOC-II Type 2 certification.
Browser Use
Make websites accessible for AI agents
Browser Use is an open-source Python library that enables AI agents to control web browsers programmatically. It translates natural language tasks into browser actions—clicking, typing, navigating, extracting data—making any website accessible to autonomous agents without APIs or integrations.
Replicate
Run open-source ML models with a cloud API
Replicate is a cloud platform that lets you run thousands of open-source machine learning models with a single API call. From image generation to speech synthesis, video creation to LLMs—access production-ready AI without managing GPUs, dependencies, or infrastructure.
Stripe
The financial operating system for autonomous economies — programmatic money movement without human intervention.
Stripe is the foundational infrastructure for the zero-human company, providing a robust API layer for every aspect of financial operations. It enables AI agents to move money, manage global tax compliance, and issue physical or virtual cards without a human CFO or finance department. By abstracting the complexity of the global banking system into a single programmable interface, Stripe allows autonomous entities to collect revenue, pay vendors, and manage capital with machine-level precision and speed.
Clerky
Startup legal automation for formation and fundraising — legal infrastructure for autonomous entities.
Clerky is the leading legal automation platform for startups, providing a programmatic way to handle company formation, hiring, and fundraising. It allows zero-human companies to maintain legal standing and execute corporate actions without human lawyers. By automating the generation and execution of legal documents, Clerky ensures that autonomous entities can operate within the legal framework of the physical world with machine-level reliability.
Vanta
Automated compliance and security monitoring — the trust layer for autonomous businesses.
Vanta automates the complex process of SOC 2, ISO 27001, and HIPAA compliance by continuously monitoring a company's infrastructure. For zero-human companies, Vanta provides a programmatic way to ensure the business remains compliant with industry standards 24/7 without human auditors. It acts as an autonomous compliance officer, identifying security gaps and evidence for audits in real-time.
Sumsub
All-in-one verification platform for KYC/KYB — programmatic identity verification for global agents.
Sumsub provides an end-to-end API for KYC, KYB, and AML screening, enabling zero-human companies to programmatically verify users and partners globally with high accuracy. It automates the entire verification process, from document checks to liveness detection, ensuring that autonomous entities can onboard customers and maintain regulatory compliance without human intervention.
Ironclad
Digital contracting platform for lifecycle management — programmatic contract execution for autonomous entities.
Ironclad's API allows for the automation of the entire contract lifecycle, ensuring that all business agreements are handled consistently and legally. For zero-human companies, Ironclad provides a programmatic way to generate, negotiate, and execute contracts without human lawyers. It acts as an autonomous legal department, managing the flow of agreements and ensuring compliance with corporate policies.
Snapshot
Off-chain voting for decentralized governance — programmatic decision-making for autonomous organizations.
Snapshot is a decentralized voting system that provides the necessary infrastructure for off-chain voting and governance. For zero-human companies operating in the Web3 space, Snapshot enables autonomous decision-making by allowing token holders to vote on proposals without incurring gas fees. It acts as a programmatic governance layer, ensuring that the organization's direction is determined by its stakeholders through a transparent and verifiable process.
Safe
Multi-sig wallet and programmable treasury — the secure foundation for autonomous capital management.
Safe (formerly Gnosis Safe) provides the infrastructure for multi-sig wallets and programmable treasury management. For zero-human companies, Safe enables secure and autonomous capital management by allowing multiple agents to co-sign transactions and manage funds through a single programmable interface. It acts as a secure vault for the organization's digital assets, ensuring that all financial actions are authorized and verifiable.
Digits
AI-native accounting and real-time financials — the autonomous CFO for zero-human companies.
Digits utilizes AI to provide real-time visibility into a company's finances, acting as an autonomous CFO by providing instant financial insights via API. For zero-human companies, Digits automates the entire accounting process, from bookkeeping to financial reporting, ensuring that the organization's financial health is always visible and verifiable without human accountants.
Ramp
Spend management and corporate card platform — programmatic expense control for autonomous agents.
Ramp provides the necessary infrastructure for spend management and corporate cards, allowing zero-human companies to issue virtual cards and manage expenses programmatically. For autonomous entities, Ramp enables agents to generate and manage their own spending limits for compute and services, ensuring that all operational costs are controlled and verifiable without human oversight.
Mercury
Banking built for startups and tech companies — the programmatic bank for autonomous economies.
Mercury provides banking services built for startups, offering a comprehensive API for domestic and international wires, virtual cards, and treasury management. For zero-human companies, Mercury enables autonomous financial operations by allowing agents to manage bank accounts and execute transactions programmatically without human intervention.
Circle
USDC and programmable money infrastructure — the bridge between traditional finance and autonomous economies.
Circle provides the infrastructure for USDC and programmable money, allowing zero-human companies to move value globally with the speed and efficiency of the internet. For autonomous entities, Circle enables programmatic payments and treasury management using stablecoins, ensuring that value can be moved and stored without the friction of traditional banking systems.
Modern Treasury
Payment operations and money movement automation — the programmatic engine for global capital movement.
Modern Treasury automates the entire lifecycle of money movement across multiple banks, ensuring seamless reconciliation without human oversight. For zero-human companies, Modern Treasury provides a programmatic way to manage complex fund flows, from onboarding users to executing payouts and reconciling transactions in real-time.
Apollo.io
Sales intelligence and engagement platform — the autonomous lead generation engine for ZHCs.
Apollo.io provides the data and CRM infrastructure needed for programmatic sales, allowing zero-human companies to identify leads and execute outreach campaigns autonomously. For autonomous entities, Apollo.io acts as an autonomous sales team, identifying potential customers and managing the entire sales funnel through a single programmable interface.
Attio
The next-generation, data-driven CRM — the programmable source of truth for autonomous relationships.
Attio is a data-driven CRM that provides the infrastructure for programmatic relationship management. For zero-human companies, Attio enables autonomous entities to track and manage interactions with customers, partners, and stakeholders through a single programmable interface. It acts as an autonomous relationship manager, ensuring that all business interactions are documented and actionable without human oversight.
11x
Autonomous AI SDR platform — the machine-led sales force for zero-human companies.
11x provides AI agents that act as Sales Development Representatives, handling outbound conversations and lead qualification autonomously. For zero-human companies, 11x enables autonomous sales operations by deploying bots that can engage with potential customers across multiple channels, ensuring that the sales funnel is always full without human SDRs.
Customer.io
Automated messaging for data-driven marketing — the programmatic voice for autonomous brands.
Customer.io triggers personalized messaging based on real-time data, allowing zero-human companies to execute marketing campaigns autonomously. For autonomous entities, Customer.io acts as an autonomous marketing department, sending the right message to the right user at the right time through a single programmable interface.
Jasper
AI content generation for marketing teams — the autonomous creative engine for ZHCs.
Jasper generates high-quality content at scale, allowing zero-human companies to maintain a consistent brand voice and execute marketing strategies autonomously. For autonomous entities, Jasper acts as an autonomous creative team, generating blog posts, social media content, and ad copy through a single programmable interface.
Intercom
AI-first customer service platform — the autonomous support department for zero-human companies.
Intercom's AI-first platform allows zero-human companies to deploy bots that handle the majority of customer inquiries autonomously, ensuring 24/7 support. For autonomous entities, Intercom acts as an autonomous support team, resolving issues and engaging with users through a single programmable interface.
Checkr
Modern, API-first background checks — the programmatic trust layer for autonomous hiring.
Checkr provides the programmatic infrastructure for background checks, allowing zero-human companies to hire and pay international talent in minutes while ensuring full compliance with local laws. For autonomous entities, Checkr acts as an autonomous compliance officer, verifying the identity and history of contractors and partners through a single programmable interface.
Deel
Global payroll and compliance for remote teams — the programmatic engine for autonomous global hiring.
Deel provides the programmatic infrastructure for global payroll and compliance, allowing zero-human companies to hire and pay international talent in minutes. For autonomous entities, Deel acts as an autonomous HR department, managing contracts, compliance, and payments for a global workforce through a single programmable interface.
Flexport
Modern freight forwarding and logistics — the programmatic engine for global supply chains.
Flexport offers APIs for global logistics and freight forwarding, enabling zero-human companies to manage complex supply chains and fulfillment processes autonomously. For autonomous entities, Flexport acts as an autonomous logistics department, managing the movement of physical goods across the globe through a single programmable interface.
Shippo
Multi-carrier shipping API for e-commerce — the programmatic engine for autonomous fulfillment.
Shippo provides a multi-carrier shipping API that enables zero-human companies to manage fulfillment and shipping processes autonomously. For autonomous entities, Shippo acts as an autonomous shipping department, comparing rates, generating labels, and tracking shipments through a single programmable interface.
Linear
Issue tracking for high-performance teams — the programmatic brain for autonomous operations.
Linear provides the programmatic interface for task tracking and project management, ensuring that all business operations are documented and organized. For zero-human companies, Linear acts as an autonomous project manager, tracking issues, managing sprints, and coordinating agent workflows through a single programmable interface.
Notion
Connected workspace for docs and databases — the programmatic knowledge base for autonomous companies.
Notion provides the programmatic interface for knowledge management and documentation, ensuring that all business operations are documented and organized. For zero-human companies, Notion acts as an autonomous knowledge base, storing documents, databases, and project plans through a single programmable interface.
Alchemy
Web3 development platform and node provider — the programmatic gateway to decentralized economies.
Alchemy provides the essential infrastructure for blockchain nodes and cross-chain data indexing, enabling zero-human companies to interact with decentralized networks. For autonomous entities, Alchemy acts as a programmatic gateway to the blockchain, providing the tools needed to deploy smart contracts, query on-chain data, and manage digital assets through a single programmable interface.
Chainlink
Decentralized oracle network for off-chain data — the programmatic bridge for smart contracts.
Chainlink acts as a decentralized oracle network, allowing smart contracts to securely interact with off-chain data feeds and web APIs. For zero-human companies, Chainlink provides a programmatic bridge between the blockchain and the physical world, ensuring that autonomous entities can access real-world data and execute actions based on external events.
The Graph
Decentralized indexing protocol for blockchain data — the programmatic search engine for Web3.
The Graph enables efficient querying of blockchain data using GraphQL, providing the necessary infrastructure for decentralized indexing. For zero-human companies, The Graph acts as a programmatic search engine for the blockchain, allowing agents to access and analyze on-chain data through a single programmable interface.
OpenZeppelin
Standard for secure smart contract development — the programmatic foundation for Web3 security.
OpenZeppelin provides the tools for secure contract development, deployment, and real-time monitoring. For zero-human companies, OpenZeppelin acts as a programmatic foundation for Web3 security, ensuring that autonomous entities can deploy and manage smart contracts with machine-level reliability and safety.
Tenderly
Full-stack blockchain development and monitoring — the programmatic observability layer for Web3.
Tenderly provides the tools for secure contract development, deployment, and real-time monitoring. For zero-human companies, Tenderly acts as a programmatic observability layer for the blockchain, ensuring that autonomous entities can debug, monitor, and optimize their smart contracts through a single programmable interface.
Moralis
Enterprise-grade Web3 data and development API — the programmatic engine for cross-chain apps.
Moralis provides the essential infrastructure for blockchain nodes and cross-chain data indexing, enabling zero-human companies to build and scale decentralized applications. For autonomous entities, Moralis acts as a programmatic engine for Web3, providing the tools needed to manage user identities, query on-chain data, and execute transactions through a single programmable interface.
Runway
AI-powered video generation and editing — the autonomous film studio for zero-human companies.
Runway allows zero-human companies to programmatically generate high-quality video content, enabling autonomous agents to maintain a consistent visual presence. For autonomous entities, Runway acts as an autonomous film studio, generating and editing videos through a single programmable interface.
Bannerbear
API for automated image and video generation — the programmatic creative engine for ZHCs.
Bannerbear allows zero-human companies to programmatically generate high-quality image and video content, enabling autonomous agents to maintain a consistent visual presence. For autonomous entities, Bannerbear acts as an autonomous creative team, generating marketing assets and social media content through a single programmable interface.
Midjourney
AI image generation via API for visual content — the autonomous artist for zero-human companies.
Midjourney provides the infrastructure for autonomous image generation, allowing zero-human companies to maintain a consistent visual presence. For autonomous entities, Midjourney acts as an autonomous artist, generating high-quality visual content through a single programmable interface.
Twilio
Global platform for SMS, Voice, and Video — the programmatic voice for autonomous agents.
Twilio provides robust APIs for SMS, voice, and video interactions, enabling zero-human companies to interact with users across multiple channels. For autonomous entities, Twilio acts as an autonomous communication department, managing global interactions through a single programmable interface.
Knock
Notification infrastructure for developers — the programmatic engine for autonomous alerts.
Knock orchestrates notifications across multiple channels, ensuring a seamless user experience for zero-human companies. For autonomous entities, Knock acts as an autonomous notification department, managing alerts and updates through a single programmable interface.
Vonage
Communication APIs for SMS and Voice — the programmatic bridge for global interactions.
Vonage provides robust APIs for SMS, voice, and video interactions, enabling zero-human companies to interact with users across multiple channels. For autonomous entities, Vonage acts as an autonomous communication department, managing global interactions through a single programmable interface.
Mixpanel
Product analytics for tracking user behavior — the programmatic brain for data-driven ZHCs.
Mixpanel provides the infrastructure for programmatic product analytics, enabling zero-human companies to track user behavior and optimize their products autonomously. For autonomous entities, Mixpanel acts as an autonomous data scientist, identifying trends and insights through a single programmable interface.
Segment
Customer data platform for data routing — the programmatic engine for data orchestration.
Segment automates data integration across various sources, ensuring that zero-human companies have a single source of truth for their customer data. For autonomous entities, Segment acts as an autonomous data engineer, routing data to the right tools through a single programmable interface.
Pinecone
Vector database for AI-driven applications — the programmatic memory for autonomous agents.
Pinecone provides a vector database crucial for AI-driven applications, enabling zero-human companies to manage and query high-dimensional data autonomously. For autonomous entities, Pinecone acts as an autonomous memory layer, providing the infrastructure needed for RAG and semantic search through a single programmable interface.
Supabase
Open-source Firebase alternative with Postgres — the programmatic backend for autonomous apps.
Supabase provides modern, API-driven database solutions, enabling zero-human companies to build and scale applications autonomously. For autonomous entities, Supabase acts as an autonomous backend, providing the infrastructure needed for databases, authentication, and storage through a single programmable interface.
Neon
Serverless Postgres database with branching — the programmatic engine for scalable data.
Neon provides modern, API-driven database solutions, enabling zero-human companies to build and scale applications autonomously. For autonomous entities, Neon acts as an autonomous database layer, providing the infrastructure needed for serverless Postgres and database branching through a single programmable interface.
Airbyte
Open-source data integration and ETL platform — the programmatic engine for data movement.
Airbyte automates data integration across various sources, ensuring that zero-human companies have a single source of truth for their data. For autonomous entities, Airbyte acts as an autonomous data engineer, moving data between tools through a single programmable interface.
HashiCorp Vault
Identity-based secrets and encryption management — the programmatic vault for autonomous security.
HashiCorp Vault provides programmatic secrets management, ensuring that zero-human companies can manage sensitive data and encryption autonomously. For autonomous entities, Vault acts as an autonomous security officer, managing access to secrets and protecting digital assets through a single programmable interface.
Doppler
Centralized secret management for developers — the programmatic engine for secure configurations.
Doppler provides programmatic secrets management, ensuring that zero-human companies can manage sensitive data and configurations autonomously. For autonomous entities, Doppler acts as an autonomous security officer, managing access to secrets across multiple environments through a single programmable interface.
Snyk
Developer security for code and dependencies — the programmatic shield for autonomous development.
Snyk ensures the security of the company's code and user identities, enabling zero-human companies to build and scale applications autonomously. For autonomous entities, Snyk acts as an autonomous security engineer, identifying and fixing vulnerabilities through a single programmable interface.
Auth0
Identity as a service for authentication — the programmatic engine for autonomous identity.
Auth0 ensures the security of the company's code and user identities, enabling zero-human companies to build and scale applications autonomously. For autonomous entities, Auth0 acts as an autonomous identity manager, providing the infrastructure needed for authentication and authorization through a single programmable interface.
0x
Liquidity aggregator for decentralized exchange — the programmatic engine for token swaps.
0x provides an API for programmatic token swaps at the best possible prices, enabling zero-human companies to manage digital assets autonomously. For autonomous entities, 0x acts as an autonomous trader, executing token swaps and managing liquidity through a single programmable interface.
Dune Analytics
Community-driven crypto data and analytics — the programmatic brain for on-chain research.
Dune Analytics enables deep on-chain research via SQL queries, providing the necessary infrastructure for zero-human companies to analyze blockchain data autonomously. For autonomous entities, Dune acts as an autonomous data scientist, identifying trends and insights through a single programmable interface.
1Password Connect
Secrets automation server for programmatic access — the secure bridge for autonomous agents.
1Password Connect provides programmatic secrets management, ensuring that zero-human companies can manage sensitive data and configurations autonomously. For autonomous entities, 1Password Connect acts as an autonomous security officer, managing access to secrets through a single programmable interface.
Anvyl
Supply chain relationship management — the programmatic engine for physical procurement.
Anvyl's API allows zero-human companies to manage supplier relationships and automate procurement workflows autonomously. For autonomous entities, Anvyl acts as an autonomous supply chain department, managing the movement of physical goods and supplier interactions through a single programmable interface.
MaxClaw
Cloud-hosted AI agent — deploy in 10 seconds, zero infrastructure
MaxClaw is the official cloud-hosted AI agent by MiniMax, built on the open-source OpenClaw framework and powered by the MiniMax M2.5 model. Deploy a persistent, intelligent assistant in 10 seconds — no servers, no Docker, no API keys to manage.
Cofounder
Agent orchestration platform — run your entire company with natural language
Cofounder orchestrates agent swarms to run companies. Natural language commands spin up agents that handle CRM, email, calendar, Notion, Linear, Slack — fully 24/7. The knowledgebase agent keeps business memory updated autonomously. This is the operational layer for zero-human companies.
Nevermined
Card payments for agents — spend mandates, scoped credentials, metering
Nevermined Pay lets an operator enroll a real card once, tokenize it, and delegate a spending mandate to an agent. The agent receives a payment capability (scoped virtual-card credential / API key), not the raw PAN. Mandates can include a budget, time window, optional per-purchase cap, transaction count, and API-key restriction; Nevermined enforces them server-side before settlement. Mid-task purchases use x402-style payment headers so the agent can buy a paid API or tool without a checkout redirect. Integrates with OpenAI, Anthropic, LangChain, CrewAI, MCP, and custom agents. Pair with the Agent Payments & Spend-Policy Control Plane playbook.
LangChain Managed Deep Agents
LangSmith managed runtime plus Connections for agent-owned vs caller-owned credentials
Managed Deep Agents (MDA) lets you define an agent as a project folder and run it on LangSmith’s managed Agent Server: durable sessions, sandboxes, Context Hub, channels, schedules, and observability. Connections are named credentials stored in the LangSmith workspace, retrieved at runtime with connections.get(). Ownership is agent-owned (shared by every caller) or caller/user-owned (resolved per person; may pause for OAuth). Ownership is fixed at create time. Public docs describe MDA as a public beta on LangSmith Cloud. Pair Connections with the Agent Payments playbook when the credential is a spend capability.
Stripe Link (Wallet for Agents)
OAuth to Link, spend request, human approval, one-time card or shared payment token
Stripe’s Link wallet for agents (built on Issuing for agents) lets a consumer grant an agent access to Link via OAuth. The agent creates a spend request with merchant, amount, and context. The person approves in Link (web or iOS/Android). Link then returns a one-time-use card or a Shared Payment Token (SPT) — scoped with controls such as amount, currency, and merchant. The agent never receives the wallet’s raw backing credentials. Stripe’s launch post states each request currently requires review before the credential is shared, with planned expansion to standing limits. Related playbooks: Agent Payments & Spend-Policy Control Plane; Harness: Permission Outside the Agent (Muse launch partnered with Stripe Link for single-use scoped cards).
METR Inspect Hawk
Inspect AI evaluations at scale on isolated infrastructure
Inspect Hawk is METR’s open-source platform for running Inspect AI evaluations on cloud infrastructure (documented AWS path: isolated Kubernetes pods, managed LLM proxy, PostgreSQL warehouse, web UI). You define tasks, agents, and models in YAML; Hawk provisions runners, streams logs, stores results, and can run Inspect Scout scans over transcripts. Inspect AI itself is the UK AI Security Institute evaluation framework; Hawk is the scale-out harness. METR (metr.org/about) evaluates frontier models’ autonomous capabilities and publishes methods; it is funded by donations and states it has not accepted funding from AI companies. Use Hawk as the acceptance-test runner in the Harness: Permission Outside the Agent playbook (behavioral evals before widening egress auto-allow).
OpenAI Agents SDK
In-app harness SDK — not the managed Agents API
The OpenAI Agents SDK is a TypeScript and Python library for building agents in your application: agent loop, tools, handoffs, guardrails, sessions, and tracing. OpenAI’s docs contrast three runtimes: Responses API (you own the loop), Agents SDK (the SDK runs the loop in your process), and Agents API (a managed harness in OpenAI’s service). Use it when your server owns deployment, tool implementations, state, and approval decisions — the same “permission outside the agent” split as the Muse/Sentinel playbook, implemented in your own harness rather than Meta’s VM.
Andon Labs Pion
Persistent, long-running agents for whole businesses — not a workflow canvas
Pion is Andon Labs’ platform of persistent, long-running agents for running whole businesses (not workflow canvases). Included tools: email, phone, banking, browser, secure bash, inventory, and staff direction via Slack. An oversight agent, Andonos, keeps operators on track. Live deployments include Andon Market (SF retail), Andon Café (Stockholm), Andon FM radio stations, and vending. Research preview waitlist is open; Andon funds selected experiments. Primaries: andonlabs.com/pion, the 2026-09-14 “Why we built Pion” post, and andonlabs.com.
Claude Managed Agents
Anthropic-managed agent harness — Agent, Environment, Session, Events
Claude Managed Agents is Anthropic’s pre-built, configurable agent harness on managed infrastructure. You define an Agent (model, system prompt, tools, MCP servers, skills), an Environment (Anthropic-managed cloud sandbox or a self-hosted sandbox on your infrastructure), then start a Session that performs a task and streams Events (user turns, tool results, status). Built-in tools: bash, file operations (read, write, edit, glob, grep), web search and fetch (optional domain allowlist/blocklist), and MCP servers. Sessions are stateful: conversation history, sandbox state, and outputs persist server-side; you can steer or interrupt mid-run. This is the managed-runtime path. Complementary: OpenAI Agents SDK (in-app harness you host, not OpenAI’s managed Agents API) and LangChain Managed Deep Agents (LangSmith-hosted Deep Agents plus Connections). Primary: https://docs.anthropic.com/en/docs/managed-agents/overview.
WorkOS Airlock
Intent-bound token and policy gate between an agent and WorkOS Pipes provider credentials
WorkOS Airlock sits between the agent and provider connections (Pipes): the agent receives an intent-bound token, and provider OAuth credentials stay behind the gateway. Airlock checks the proposed action and content against policy before the call proceeds (allow, deny, or human-approval). Declared intent does not grant a policy exception. WorkOS lists Airlock as early access (product page and 23 Sep 2026 guide). At init() on 7 Oct 2026 it previewed Airlock with Atlas and Passport and said all three are previews rolling out over the coming weeks and months; neither post gives pricing. Enforce rules handle deterministic checks (endpoint, method, request-body value); runtime rules go to a governing LLM inside Airlock that reads the policy. See https://workos.com/blog/init-2026-recap and https://workos.com/airlock. Unstealable credentials are not enough if action and content are unconstrained. See https://workos.com/blog/workos-airlock and https://workos.com/blog/mcp-authorization-failures.
WorkOS AuthKit
AuthKit as MCP authorization server plus Agent Registration for programmatic identity
WorkOS AuthKit can act as a spec-compatible OAuth authorization server for an MCP resource server: Client ID Metadata Document (CIMD), resource indicators (RFC 8707), protected-resource metadata, and audience-bound access tokens. Your MCP server verifies tokens issued for itself and points clients at AuthKit via standardized metadata. Agent Registration is a separate surface for programmatic clients: anonymous or service_auth registration, an optional claim ceremony that binds an agent to a user, then assertion exchange for a short-lived access token or API key with act/delegation claims. Agent Registration must be enabled per environment (contact the WorkOS account team if it is not available). Cross App Access is marked early access on the MCP docs — do not treat it as GA. Complementary to /playbooks/agent-credential-architectures/ and /playbooks/agent-identity-kya/. Primaries: https://workos.com/docs/authkit/mcp and https://workos.com/docs/authkit/agent-auth.
Earendil Pi
Pi 1.0 ships MCP in core; experimental Pi Durable harness over pluggable storage
Pi is a minimal agent harness. At 1.0 it ships MCP in core via its codemode sandbox. Earendil’s stated reason: MCP should behave more like OpenAPI, with structured returns and discovery by documentation, not by dumping tools into context. The same cycle introduced experimental MIT-licensed Pi Durable: a harness over pluggable storage (memory, SQLite, JSONL, plus a conformance suite) for long-running agents that survive crashes, can be reached from several surfaces, and let several humans steer the same agent. About 15,000 lines so an agent can read it. Pick harness storage you control, keep tool returns structured, and treat durable sessions as long-lived principals for revocation. Durable storage must sit outside the agent’s write permissions — see /playbooks/harness-permission-outside-agent/. Primaries: https://earendil.com/posts/you-said-no-mcp/ ; https://earendil.com/posts/pi-durable/.
Cloudflare Agents SDK
Durable agent runtime on Workers / Durable Objects
Cloudflare’s durable agent runtime on Workers / Durable Objects: each agent session gets a durable identity, local SQL storage, real-time connections (WebSockets), scheduling, fibers, and recoverable execution. Operators compose communication channels (chat, voice, email, Slack, webhooks), an agent harness (Project Think or a custom loop), the Agents SDK runtime (`Agent` class, state, sessions, routing), and tools (Browser, Sandbox, AI Search, MCP, Payments). Starter via `npx create-cloudflare@latest --template cloudflare/agents-starter`; default Workers AI, swappable to OpenAI / Anthropic / Gemini. Stack slot: edge-hosted durable agent runtime — complementary to Anthropic Managed Agents, OpenAI Agents SDK, and LangChain Managed Deep Agents. Primary: https://developers.cloudflare.com/agents/.
MCP TypeScript SDK
Official TypeScript libraries for building MCP servers and clients
The MCP TypeScript SDK is the official Model Context Protocol implementation for TypeScript, maintained in the modelcontextprotocol GitHub organization. Version 2 is the stable release line and implements the 2026-07-28 MCP specification. It ships as split packages: @modelcontextprotocol/server for building servers and @modelcontextprotocol/client for building clients, plus thin optional middleware packages for Node.js HTTP, Express, Fastify and Hono. It runs on Node.js, Bun and Deno, and tool and prompt schemas use Standard Schema, so Zod v4, Valibot or ArkType all work. The v1 line (@modelcontextprotocol/sdk) lives on a separate branch and, per the repository README, keeps getting bug fixes and security updates for at least six months after the v2 release on 27 Jul 2026. On 7 Oct 2026 npm listed 2.3.1 for the v2 packages and 1.32.1 for v1.
Inspect AI
Open-source framework for LLM and agent evaluations
Inspect is an open-source framework for frontier AI evaluations, developed by the UK AI Security Institute and Meridian Labs. Evaluations are built from composable datasets, agents, tools and scorers, and the project ships more than 200 pre-built evaluations that run on any supported model. It supports agent evaluations with built-in agents, multi-agent primitives and the ability to run external agents such as Claude Code, Codex CLI and Gemini CLI. Tool calling covers custom and MCP tools plus built-in bash, Python, text editing, web search, web browsing and computer tools, and untrusted model code can run in sandboxes on Docker, Kubernetes, Modal, Proxmox, Vagrant and others. The Inspect View web tool and a VS Code extension help monitor and debug runs. Install with pip install inspect-ai.
Browserbase
Cloud browsers, search and fetch APIs for agents that act on the web
Browserbase is a platform for running agents that browse and act on websites. One API key covers cloud browsers that agents control to navigate pages and fill forms, a Search API for finding relevant sites, a Fetch API that returns a page as HTML, JSON or markdown, a runtime for deploying agents, and an agent identity layer for logins and CAPTCHAs. Sessions come with logs, live view and session replay. Browserbase works with Playwright, Puppeteer and Selenium, and maintains Stagehand, an open-source (MIT) browser automation framework whose act, extract and observe primitives sit on top of Playwright-style control.
x402 SDKs
Open-source TypeScript, Python and Go SDKs for charging and paying per request over HTTP 402
x402 is an open standard for internet-native payments built on the HTTP 402 Payment Required status code. A resource server answers a request with the payment it requires; the client sends a signed payment payload; the server verifies it, either locally or through a facilitator, and settles it. The reference SDKs live in the x402 Foundation's GitHub repository: TypeScript packages on npm under @x402 (core, chain packages, client wrappers for fetch and axios, and server middleware for Express, Fastify, Hono and Next), a Python package (x402 on PyPI) and a Go module. Version state as of 21 October 2026: the latest npm releases of the @x402 packages are 2.28.0 (29 September 2026), the Go module is v2.28.0, and the Python package is 2.25.0 (29 September 2026). A fix merged on 13 October 2026 (pull request #3796) closes a payment bypass where HEAD requests could reach paid GET handlers without payment verification, in the TypeScript, Go and Python route matchers; no published release includes it yet. The unscoped npm package named x402 is the older version 1 line and is still at 1.2.0 (16 April 2026); install the scoped @x402 packages for version 2.
AWS Agent Registry
A managed catalog where teams publish, approve and discover MCP servers, agents and skills
AWS Agent Registry is a fully managed discovery service in Amazon Bedrock AgentCore. Teams publish MCP servers, tools, agents, agent skills and custom resources as records in a registry; a curator approves or rejects each record before it becomes discoverable, and can deprecate it later. People and agents find records through hybrid semantic and keyword search, catalog browsing, or the registry's own remote MCP endpoint. AWS says the registry validates MCP and agent records against their protocol schemas. Access is controlled with AWS IAM credentials or JSON Web Tokens from a corporate identity provider, API calls are logged in AWS CloudTrail, and Amazon EventBridge can notify when a record is submitted for approval.