Browse AI tools by what you need to get done.
Tools by goal26 step-by-step guides→All categoriesBy goal
An AI feature that works in a demo is a small part of the job. The demo is one prompt, one model and five friendly inputs. The product is the same thing running against thousands of unpredictable users, with your own data behind it, a bill that scales with usage and a need to know when answers get worse.
So the stack has layers. A model, reached through an API or hosted yourself. Prompts that are versioned like code. Retrieval and memory, so the model answers from your content instead of from guesswork. Possibly an agent loop with tools it may call. Ordinary application code around all of it, then tests, evaluation and a place to deploy.
Coding agents now write a large share of that application code, and app builders can produce a working front end from a description. What they do not settle is the engineering judgment: which model is good enough for the task at an acceptable cost, how to measure quality before users do, what data may leave your system, and what happens when the model is confidently wrong. Models and their terms also change often, so design the app so that a model can be swapped without a rewrite.
Who this is for Developers, technical founders and small product teams adding a language-model feature to a product or building a new app around one, who want a map of the tooling before choosing a stack.
A common first pass for a prototype: compare models on your own prompts, write the code with an agent, add tracing early, then deploy. The picks follow that order, not a ranking.
Turn your rough task description into a structured prompt, then test it on twenty real inputs.
Describe the smallest version of the feature and review the plan before accepting any changes.
Turn on tracing from the first prototype so every bad answer becomes a test case.
You decide which model or models the app will call, weighing output quality on your own task, speed, context size, cost per request, privacy requirements and whether you need a hosted API or an open model you run yourself.
Offers its language models through a token-billed API that accepts OpenAI-style and Anthropic-style requests, with JSON output and tool calls.
AI research company offering a chat assistant, a desktop harness and a token-billed API for its language models.
Runs open-source models inside secure hardware enclaves and exposes them through an OpenAI-compatible API, for apps that handle sensitive data.
Private AI cloud that runs models inside secure hardware enclaves, offered as a chat app, an inference API and containers.
Before writing application code you work out the instructions: system prompt, examples, output format. You try them against real inputs, including awkward ones, and keep each version so that changes can be compared.
Turns a short, rough request into a clear, structured prompt and stores every prompt in a searchable library for reuse.
Prompt generator that turns a rough request into a structured prompt for ChatGPT, Claude, Gemini and other models.
Analyzes a prompt for strengths and weaknesses, rewrites it in rounds and records each version so earlier ones can be recovered.
Prompt workspace for generating, analysing, optimising and reverse-engineering prompts for text, image and video models.
You connect the app to models in a way that survives production: keys kept server-side, retries and fallbacks when a provider fails, rate limits, spend tracking, and GPU hosting if you serve an open model yourself.
Its AI Gateway routes requests to many AI models through one service, with fallback, load balancing, retries and spend monitoring.
Cloud platform for deploying web apps and AI agents, with an AI gateway, sandboxes, workflows and a global CDN.
Serves LLMs and media models through endpoints on autoscaling GPU containers that scale to zero when idle.
Serverless cloud platform for running AI inference, training, sandboxes and batch jobs on GPUs.
Documents how to call its models from application code and coding agents, with thinking mode that can be switched off per request.
AI research company offering a chat assistant, a desktop harness and a token-billed API for its language models.
Most useful apps answer from specific content: your documentation, a customer's files, the live web, past conversations. You ingest and index that content, retrieve the relevant pieces per request and pass them to the model with sources.
Searches, scrapes and crawls websites through an API and returns Markdown or structured JSON suited to retrieval pipelines.
Web data API that searches, scrapes, crawls and interacts with websites and returns clean output for AI applications.
Stores and indexes embeddings in a Postgres database so the application can run similarity search with its chosen models.
Postgres development platform with auth, instant APIs, edge functions, realtime, storage and vector embeddings.
Ingests docs, tickets and wikis and offers a Retrieval API that returns relevant chunks with source URLs for a query.
Ingestion and retrieval platform that turns company docs, tickets and wikis into AI agents and a search API.
Builds a per-user graph of dated facts from data and conversations and returns the context a task needs in one call.
Context and memory layer for AI agents that builds governed knowledge graphs from business data, documents and conversations.
If the app must take actions, not only answer, you build an agent loop: the model decides which tool to call, your code runs it and returns the result. You define the tools, their permissions and when a human must confirm.
Provides the open source langchain, langgraph and deepagents libraries for writing agents at different levels of control.
Open source agent frameworks plus the LangSmith platform for tracing, evaluating and deploying AI agents.
Publishes ordinary Python functions as MCP tools, generating their schemas and handling validation, transport and authentication.
Python framework for building Model Context Protocol servers, clients and interactive apps.
Connects an agent to MCP servers from a catalog and manages OAuth, credentials and sessions for those connections.
Registry and connection layer that links AI agents to MCP servers with managed OAuth and credentials.
Around the model sits a normal application: API routes, data models, authentication, background jobs, a user interface. You write, review and test this code, increasingly by delegating tasks to a coding agent and reviewing its changes.
Takes a written development task, explores the codebase, proposes a plan and makes the changes for you to accept or revise.
AI coding agent and editor that plans, writes and reviews code on the desktop, in the terminal and in the cloud.
Plans programming tasks, edits files, runs tests and opens pull requests inside your repository, asking before it modifies files or runs commands.
Anthropic's AI coding agent that plans, edits code, runs tests and opens pull requests from your terminal or IDE.
Runs as an open source coding agent in the terminal, desktop or IDE and connects to many model providers, including locally hosted models.
Open source AI coding agent for the terminal, desktop and IDE that works with models from many providers.
Reviews every pull request with line-level comments and committable suggestions, which helps when much of the code is agent-written.
AI code review that comments on every pull request, with PR triage, change explanations and security monitoring.
Some apps need more than text: spoken replies, transcription, real-time voice conversation, generated images or clips. These are separate models with their own APIs, latency budgets and usage rules.
Supplies realtime text-to-speech, speech-to-text and a speech-to-speech API that streams audio over one WebSocket or WebRTC link.
Realtime voice AI platform offering text-to-speech, speech-to-text, a speech-to-speech API and LLM routing.
Provides WebRTC infrastructure and maintains Pipecat, an open-source framework for orchestrating real-time voice AI agents.
Developer platform for real-time voice, video and AI agents, built on WebRTC infrastructure and the open-source Pipecat framework.
Exposes its video and image generation models through a developer API with documentation and a model router.
AI platform for generating and editing video, images and audio, with a developer API and world models.
You build a test set from real inputs, score outputs before each change, and after launch trace every request so that failures, cost and latency are visible. Human review covers what automatic scoring cannot judge.
Its LangSmith platform records every step of an agent run, builds datasets from real traces and runs online and offline evaluations.
Open source agent frameworks plus the LangSmith platform for tracing, evaluating and deploying AI agents.
Sends AI evaluators against chat and voice agents to check for hallucinations, bias, toxicity and compliance issues before release.
Software testing cloud, formerly LambdaTest, with AI agents that plan, write and run tests for web, mobile and AI apps.
Its AI observability tracks the events generated by an application's model calls, next to session replays and error tracking for the rest of the app.
Product data platform combining analytics, session replay, feature flags, error tracking and a warehouse, with AI on top.
The app needs a front end, user accounts, hosting, a domain and a release process with previews and rollback. For a prototype or internal tool this can be generated, while for a product it is deliberate engineering.
Builds and deploys a project on each Git push, giving every change its own preview URL and allowing instant rollback.
Cloud platform for deploying web apps and AI agents, with an AI gateway, sandboxes, workflows and a global CDN.
Builds a web app from a plain-language description with a live preview, and handles hosting and authentication on the platform.
AI app and website builder that turns plain-language descriptions into working full-stack software.
Generates a web application from a prompt, syncs the code to GitHub and publishes it to production.
Vercel's AI app builder that turns prompts into full-stack web apps and publishes them.
Handles sign-ups and logins and exposes database tables through REST endpoints, so the app has a backend without custom server code.
Postgres development platform with auth, instant APIs, edge functions, realtime, storage and vector embeddings.
Start with a hosted API unless you have a clear reason not to. It is faster to build on and needs no GPU operations. Running an open model yourself makes sense when data must stay in your environment, when you need to fine-tune, or when volume is high enough that owning the compute is cheaper. Many teams prototype on an API and revisit the choice once usage is known.
Often not. A single model call with good instructions and retrieved context covers many features: summarizing, classifying, extracting, answering from documents. A framework earns its place when the model must choose between tools, run several steps, keep state or recover from failures. Adding one early also adds concepts to debug, so begin simple and adopt it when the plain approach stops being enough.
A prototype or internal tool, yes. App builders in this guide generate a working web app with login and hosting from a description. A product that handles customer data and payments is different: someone needs to understand security, cost and failure handling. A sensible path is to prototype without code, test with real users, then bring in a developer before scaling.
Collect a set of real inputs, write down what a correct answer looks like for each and score the system against it before every change. Include difficult and adversarial cases. Combine automatic checks with human review of a sample. Launch to a small group with tracing on, read the failures, and widen access as the failure rate becomes acceptable for your use.
Tools are placed in a stage by an editor, from each tool’s own profile. The list grows as more profiles are published. All guides
