tools i use

agent inboxes

give an agent its own mailbox to send, receive, and thread.

• agentmail my current one. agents get real inboxes via pure api, no human mail client.
• ai inbx same idea, leaning hard on no oauth, no app passwords, one api key, plus auto context-aware threading.
• robotomail agent self-signup with a single post, webhooks, sse, or polling for inbound, and a cli for debugging.
• infraforge dedicated ip and domain per agent, built for deliverability and cold outreach instead of a shared pool.

browser & web

let an agent actually drive a browser and read the live web.

• browserbase cloud headless browsers at scale. stagehand is the ai-native automation layer on top.
• steel open-source headless browser api, same lane as browserbase.
• firecrawl turns any site into clean, llm-ready markdown. the default scrape-this-for-the-agent tool.
• playwright mcp microsoft's mcp server so an agent drives playwright with zero glue code.
• jina reader turns any web page or pdf into clean, llm-ready text, and does web search built for models.

code sandboxes

somewhere safe for an agent to run generated code.

• e2b the de facto secure code-execution sandbox for agents. open-source, spins up disposable vms.
• daytona / modal adjacent runtimes for sandboxed or gpu-backed agent compute.

tool integration & auth

one layer that hands the agent hundreds of pre-wired actions.

• composio 150+ tools like gmail, slack, and github with managed oauth and token lifecycle, so the agent calls structured tools instead of fighting each api's login.
• arcade same category, strong on per-user auth so the agent acts as a specific user safely.
• stackone / toolhouse unified-api plays that map one schema across many saas tools.

memory

persistent state so the agent remembers across runs.

• mem0 drop-in memory layer. the agent reads and writes long-term memory via api.
• zep similar, with temporal knowledge-graph memory.

search & retrieval

web knowledge built for llms, not humans.

• exa neural search api designed for agents. semantic, returns content not just links.
• tavily search api purpose-built for rag and agent pipelines.

payments & identity

the newest, fastest-moving layer, mostly 2025 and 2026.

• stripe agentic commerce agents can hold a wallet, get shared payment tokens, and transact via the agentic commerce protocol. the agent toolkit drops stripe actions straight into agent frameworks.
• coinbase agentkit embedded and server crypto wallets so agents transact onchain with usdc on base.
• x402 http-native pay-per-call protocol. an agent literally pays per api request inline.
• skyfire micropayments and identity for agents, with a roughly 10-minute implementation.
• world agentkit proof of human. biometric world id plus x402 so a transaction proves a real human authorized the agent.
• visa / mastercard agent pay the card networks' agent-authentication rails, scaling through 2026. visa trusted agent protocol and mastercard agent pay.

languages

the languages the stack is written in.

• ts typescript adds types to javascript so mistakes get caught before the app runs.
• python the language of ai and data, plus a solid backend and scripting workhorse.
• go compiled language for fast apis, networking, and single-binary cli tools.
• rust memory-safe systems language for performance-critical code and wasm.
• sql the query language for relational databases, the thing you actually ask your data in.

frameworks

the base your whole app runs on, both what people see and what happens behind the scenes.

• next.js full-stack react framework with routing, server rendering, and api routes. the default base for the stack.
• react component-based library for building uis out of reusable pieces.

ui & styling

how the app looks, edited directly by the agent with no design system to learn.

• tailwind utility-first css the agent edits straight in your markup.
• shadcn accessible react components you copy in and own, so the agent restyles them directly.
• daisyui tailwind component library with ready-made class-based ui like buttons, cards, and modals.
• motion animation library for react, formerly framer motion, for transitions and gestures.
• aceternity ui copy-paste animated react sections built on tailwind and motion, for flashy marketing pages.
• magic ui library of animated components and effects that pair with shadcn and tailwind.
• radix ui unstyled, accessible react primitives that shadcn is built on top of.
• tremor react components for dashboards, charts, and data-heavy uis.
• v0 vercel's ai that generates react and tailwind ui from a prompt.

databases

where your app stores and loads data like users, posts, sessions, and settings.

• supabase postgres plus auth, storage, and realtime in one sdk, one dashboard, one key. agents pick it because it collapses four services into one integration.
• neon serverless postgres with git-style branching. lean when you only want the db.
• upstash serverless redis and queues. http-based so it works anywhere with just a token.
• mongodb + mongoose document database, with mongoose as the schema and modeling layer on top.

auth

signing people up, logging them in, and knowing who they are.

• clerk install the sdk, drop in pre-built react components, instant auth. near-zero config.
• nextauth v5 self-hosted auth for next.js with google and email sign-in built in.

email

sending email from your app.

• resend transactional email with react email templates. one resend_api_key and send.

payments & billing

taking money, from raw payments to fully managed billing.

• stripe payments, mature enough that the agent already knows the integration cold.
• polar / lemon squeezy merchant-of-record billing if you want stripe-level ease but want them to handle the tax and compliance headache.

file uploads

letting users upload files like images and documents.

• uploadthing file uploads in a few lines. pairs with the next.js stack.

file storage

where the files themselves actually live.

• cloudflare r2 s3-compatible object storage with no egress fees, good for images and large files.
• aws s3 the original object storage standard that almost everything integrates with.
• tigris globally distributed, s3-compatible storage that puts files near your users.

background jobs

running work outside the request, on a schedule or in the background.

• trigger.dev / inngest background jobs and durable workflows without managing a queue.

analytics & monitoring

seeing what users do and catching what breaks.

• posthog analytics, flags, and session replay in one snippet.
• sentry error tracking in one wrapper.

support

talking to your users and handling support.

• crisp drop-in live chat and support widget you add with one snippet.
• intercom customer messaging suite, now ai-first with its fin support agent.
• plain modern support platform built for technical and developer-facing teams.
• chatwoot open-source live chat and support inbox you can self-host.

deployment

putting your app online so anyone can visit it.

• vercel push and it deploys. the agent does not touch infra.

ai

the model that writes, reasons, and powers anything smart in your app.

• claude anthropic's frontier llm. best for coding, writing, and agentic workflows that need real reasoning.

agent frameworks

the libraries you actually build agents with.

• vercel ai sdk typescript toolkit for streaming, tool calls, and agents across model providers.
• mastra typescript framework for building agents with workflows, memory, and tools.
• langchain the big framework for chaining models, tools, and retrieval into agents.

vector db & rag

storing embeddings so the app can search by meaning.

• pinecone managed vector database for semantic search and retrieval at scale.
• qdrant open-source vector database with fast filtering, self-host or cloud.
• turbopuffer serverless vector and full-text search built on object storage, cheap at scale.

llm observability

seeing what your models and agents actually did.

• langfuse open-source tracing, evals, and analytics for llm and agent apps.
• helicone drop-in proxy that logs, caches, and monitors your llm calls.
• braintrust evals and observability for shipping llm features with confidence.