09 · agent-services · 15 systems
Agent servicesThe substrate agents can't self-provision — payments, identity, inboxes, sandboxes, search, memory.
Agent services are purpose-built primitives whose customer is the agent (or its builder): ways to pay (x402, AP2, AgentCard, Skyfire), ways to authenticate (Keycard, Arcade, Scalekit), places to exist (AgentMail inboxes, E2B sandboxes, Browserbase browsers), and senses (Exa search, Mem0 memory). Where bridges give agents protocols, services give them accounts — metered, billable capabilities the agent can't synthesize.
Systems in this category
AgentMail
AgentMail · agent-services
AgentMail is email infrastructure for agents — API-created inboxes, sending, and parsing that give an agent a durable, interoperable address for agent-to-agent and agent-to-human communication.
x402
Coinbase · agent-services
x402 is Coinbase's open payment protocol — it revives HTTP 402 Payment Required so an agent that hits a paid resource can settle programmatically in stablecoins (USDC on Base) and get the resource, no human checkout step.
Agent Payments Protocol
Google · agent-services
Agent Payments Protocol (AP2) is Google's open spec for agent-initiated payments — signed mandates that prove a user authorized a purchase, riding existing card networks and payment rails instead of inventing new ones.
Skyfire
Skyfire · agent-services
Skyfire is a payments and identity network for the agent economy — it gives agents a wallet and verifiable 'KYA' (know-your-agent) credentials so they can pay APIs, MCP servers, and each other.
Payman
Payman · agent-services
Payman is agent payments with the approval layer built in — wallets, policies, and human-in-the-loop sign-off so an agent can spend money it can't lose.
Nevermined
Nevermined · agent-services
Nevermined is agentic-commerce infrastructure — payments, subscriptions, and metering built for agents buying from agents, with agent-to-agent payment plans and credits.
AgentCard
AgentCard · agent-services
AgentCard issues virtual payment cards for agents from a CLI — a one-line install provisions the agent's card, email inbox, and x402 wallet so it can check out anywhere a card is accepted.
Keycard
Keycard · agent-services
Keycard is identity and access for autonomous agents — issues agents their own credentials, applies policy to what each agent may touch, and audits it like an employee's badge.

Arcade
Arcade · agent-services
Arcade is auth and tools for agent builders — a managed way for agents to act on a user's behalf (email, calendar, Slack, GitHub) with scoped OAuth tokens, plus a tool-calling SDK.
Scalekit
Scalekit · agent-services
Scalekit is the auth stack for agent apps — MCP-server auth, agent OAuth flows, and SSO/SCIM for the humans, so one platform covers who the user is and what the agent may do.
Composio
Composio · agent-services
Composio is a tool-and-integration platform for agents — hundreds of SaaS tools (Gmail, Slack, GitHub, Linear) exposed to any agent framework with managed auth and consistent schemas.
E2B
E2B · agent-services
E2B provides secure sandboxes for agent-generated code — instant isolated VMs where an agent can run untrusted code, install packages, and keep files, widely used as the compute layer behind AI apps.
Browserbase
Browserbase · agent-services
Browserbase is browsers-as-a-service for web agents — hosted headless browsers with stealth, proxies, sessions, and replay, the runtime underneath its own Stagehand framework and third-party agents.
Exa
Exa · agent-services
Exa is a search engine built for agents — an index queried by meaning (neural + keyword) with APIs that return clean page contents instead of a page of links.

Mem0
Mem0 · agent-services
Mem0 is the memory layer for agents — extract, store, and retrieve user-and-agent memories across sessions, as open-source SDK or hosted API.
Who it's for
Builders shipping agents that must transact, authenticate, or persist beyond a chat loop — agents that buy things, hold credentials, own an inbox, or run untrusted code somewhere safe.
Tradeoffs
Each service is a small surface and easy to swap; the risk is lock-in on the regulated pieces — payments and identity are the parts auditors look at. The 402-style payment standards are converging fast; card rails remain the interoperable fallback.
Frequently asked
What is an agent service?
Agent services are purpose-built primitives whose customer is the agent (or its builder): ways to pay (x402, AP2, AgentCard, Skyfire), ways to authenticate (Keycard, Arcade, Scalekit), places to exist (AgentMail inboxes, E2B sandboxes, Browserbase browsers), and senses (Exa search, Mem0 memory). Where bridges give agents protocols, services give them accounts — metered, billable capabilities the agent can't synthesize.
Who should use an agent service?
Builders shipping agents that must transact, authenticate, or persist beyond a chat loop — agents that buy things, hold credentials, own an inbox, or run untrusted code somewhere safe.
What are the tradeoffs of agent services?
Each service is a small surface and easy to swap; the risk is lock-in on the regulated pieces — payments and identity are the parts auditors look at. The 402-style payment standards are converging fast; card rails remain the interoperable fallback.
Which agent services are in the catalog?
15 as of 2026-09-25: AgentMail (AgentMail), x402 (Coinbase), Agent Payments Protocol (Google), Skyfire (Skyfire), Payman (Payman), Nevermined (Nevermined), AgentCard (AgentCard), Keycard (Keycard), Arcade (Arcade), Scalekit (Scalekit), Composio (Composio), E2B (E2B), Browserbase (Browserbase), Exa (Exa), Mem0 (Mem0).
What counts as an agent service?
A metered product whose user is an agent or its operator, providing a capability the agent can't produce itself — money movement, identity and credentials, an email address, a sandbox, a search index, a memory store. Distinct from bridges (protocols and transports): services are products with accounts and billing.
What is HTTP 402 / agentic payments?
402 Payment Required is the long-dormant HTTP status code being revived for machine-native payment — an agent hits a paid resource and settles programmatically instead of a human entering card details. Coinbase's x402 rides crypto rails; Google's AP2 defines signed mandates over card networks and payment rails.
How do agents authenticate as themselves?
Two models coexist: delegated auth (Arcade, Scalekit) issues agents scoped tokens to act on a user's behalf; agent-first identity (Keycard) gives the agent its own credentials and policy. Most production systems need the first today and will want the second as agents act more independently.