AI assistants and agents

All 115

Alphabetical. Checked on 2026-09-04; anything that did not answer says so.

• 10Web
  Generates and hosts WordPress sites, with the build and the hosting sold as one product.
  WordPress without meeting WordPress.
• Agent Shadow Brain the page returns 404
  Listed as a background analysis agent. The repository returns 404.
  Gone: the repository is not there.
• AgentsMesh
  Assigns work to a group of agents and tracks it, treating them as a team roster rather than as one assistant.
  Org chart as a product feature.
• AI SDK by Vercel
  A TypeScript toolkit that puts one interface in front of many model providers, including streaming and tool calls.
  One API, so the provider becomes a config value.
• Aide
  An agent that spends more compute at inference time on harder engineering problems rather than answering immediately.
  Buys accuracy with time.
• Aider
  Pair programming in the terminal. Reads the repository, proposes edits, and commits them to git with a message.
  Every change lands as a commit you can revert.
• Amazon CodeWhisperer
  Amazon's original completion assistant. The name is retired and the address now serves Amazon Q Developer.
  Renamed, not retired.
• Amazon Q Developer
  Amazon's coding assistant, integrated with AWS accounts and services rather than with the editor alone.
  An assistant that knows which account it is standing in.
• Amp (Sourcegraph)
  Sourcegraph's agent and development environment, including unsupervised runs in hosted sandboxes.
  Lets the agent off the leash on purpose.
• Augment Code
  An agent platform aimed at the organisation rather than the individual engineer, which is where it draws its dividing line.
  Argues the gap is organisational, not personal.
• AutoGen
  Microsoft's framework for multi-agent conversations, where agents pass work between each other under a defined protocol.
  Made agents talking to agents a research programme.
• base44
  Describes an application in plain language and produces a working one, hosting included.
  Aimed at people who will never open the repository.
• Blackbox AI
  Routes many models through one endpoint, or runs an open-weight model as a dedicated deployment.
  Model choice as a runtime decision.
• Bolt.diy
  The open version of Bolt: prompt, run and edit full-stack applications in the browser against any model you choose.
  Same idea, your model, your fork.
• Bolt.new
  Builds and deploys full web applications from a prompt, running the whole toolchain in the browser.
  No local environment at any point.
• brood-box
  Runs coding agents inside hardware-isolated microVMs, so an agent with shell access cannot reach the host.
  Assumes the agent will do something you did not authorise.
• Builder.ai the certificate does not validate
  Assembled custom software from reusable parts. The site fails its TLS handshake and does not load.
  Unreachable: the connection is refused at the handshake.
• Builder.io Fusion
  Puts designers, engineers and marketers on one codebase, with visual edits landing as real code.
  Removes the handoff by removing the second artefact.
• Capacity
  Asks questions about an idea, then builds and hosts the application including accounts, payments and a domain.
  Interviews you instead of taking a prompt.
• Claude Code
  Anthropic's agentic coding tool. Runs in the terminal with access to the repository, the shell and the test suite.
  An agent that can run the tests it just broke.
• Claude Desktop
  Claude as a desktop application, with MCP connections to local tools and files.
  Where MCP stopped being a specification and became a menu.
• Cline
  An autonomous coding agent that ships as a VS Code extension, a CLI and an SDK, with read and write access to the working tree.
  The agent moved out of the chat panel and into the file system.
• CodeGeeX
  An open multilingual code model and editor extension, developed largely outside the Western vendor set.
  A reminder that this field is not only American.
• Codeium
  A free completion tool that became Windsurf and was then acquired. The address now serves Devin Desktop.
  Absorbed: the URL now belongs to somebody else.
• Codename Goose
  Block's open-source agent that installs, executes, edits and tests, against whatever model you point it at.
  Extensible where most agents are opinionated.
• CodeParrot
  A VS Code assistant narrowed to front-end components rather than to all code.
  Picked one job instead of every job.
• CodeSquire the domain does not resolve
  A browser extension bringing completion to notebooks and query editors. The domain no longer resolves.
  Gone: the domain does not resolve.
• Codex Infinity
  Runs long autonomous coding sessions in the cloud with orchestration across them.
  Sells session length as the feature.
• Cody (Sourcegraph)
  Sourcegraph's assistant, built on top of a code search index rather than on the open file.
  Retrieval first, generation second.
• Coframe
  Runs continuous optimisation on a live site, generating and testing interface variants without a person writing them.
  A/B testing where the model writes the B.
• Continue
  An open-source assistant that let you point the editor at any model, local or hosted. Acquired by Cursor.
  The bring-your-own-model bet, bought by the competition.
• Copilot Workspace
  GitHub's agentic environment, taking an issue through plan, edit and pull request inside the platform.
  Starts at the issue, not at the file.
• CopilotKit
  React components for putting agent interfaces into a product, plus the protocol underneath them.
  The front end of agents, which most frameworks skip.
• Create.xyz
  Turns a description into a working application, aimed at people who will not open a terminal.
  The prompt is the entire interface.
• CrewAI
  A framework for agents with defined roles that hand work between each other, plus a runtime for operating them.
  Modelled on a team, down to the job titles.
• Cursor
  A fork of VS Code rebuilt so the model is a first-class editor, not a sidebar. Now positions itself as an agent you hand tasks to.
  The editor stopped completing lines and started accepting assignments.
• Databutton
  Listed as an app-building agent. The address now serves Riff, a procurement and logistics product.
  The domain moved on to a different business.
• Deepsite
  Builds and deploys websites from a description, hosted on open model infrastructure.
  Site generation on open weights.
• Devin AI
  An autonomous software engineer running parallel cloud agents. The company that set expectations for this category, and absorbed two competitors.
  Defined the category, then bought the competition.
• Devon the connection timed out
  An open agent for autonomous coding. The site timed out on every attempt.
  Unreachable: the connection timed out.
• Durable
  Generates a small business website with marketing and customer tools attached.
  Aimed at the trade, not the developer.
• Dyad.sh
  An open-source desktop app builder that runs locally against any model, with the code staying yours.
  Local, open, and explicit about lock-in.
• Factory
  Agents, called Droids, that carry work across coding, testing and deployment rather than stopping at the diff.
  Names its agents, which tells you the intended relationship.
• Figma AI
  Model features inside Figma for search, generation and the repetitive parts of design work.
  Aimed at the busywork, not the concept.
• Fine the site refused an automated request
  An agent that works from stated requirements and iterates on its own output. The site returned a payment-required error when checked.
  Requirements in, pull requests out.
• FlutterFlow
  Visual development for Flutter, producing real Dart that can be exported and taken away.
  Visual building that gives the source back.
• Forge
  Generates full-stack applications from a description, with verification in the loop rather than a debugging cycle.
  Claims to check its own output before handing it over.
• Framer AI
  Generates editable pages inside Framer, with the design canvas and publishing already there.
  Generation into a design tool, not into a repository.
• Freepik the site refused an automated request
  Stock assets and generative image tools for design work. The site refused an automated request.
  Stock imagery, now generated as well as licensed.
• Gemini CLI
  Google's open-source terminal agent, putting Gemini in the shell with tool access.
  Google's entry in the terminal agent race, open sourced.
• GitHub Copilot
  The first AI completion tool to reach general availability, now spanning chat, review and agents across editors and GitHub itself.
  The one that made autocomplete a product category.
• Glide
  Builds internal business software from a spreadsheet, a file or a prompt, aimed at operations teams rather than engineers.
  The spreadsheet becomes the application.
• Google Antigravity
  Google's agent-first development platform, with browser control and multi-agent orchestration.
  Google's second answer, after Code Assist.
• Google Code Assist (Gemini)
  Gemini-backed completion and chat wired into Google Cloud's own services and consoles.
  An assistant that already knows your project ID.
• GPT Engineer
  An early experiment in generating whole codebases from a prompt. It became Lovable, and says so on its own repository.
  The prototype that turned into a company.
• Horizon UI the site refused an automated request
  A component boilerplate with generative UI pieces included. The site returned a payment-required error.
  A starting template, not a framework.
• JetBrains AI
  AI assistance built into the JetBrains IDEs, using the static analysis and refactoring engine those editors already had.
  A model bolted onto an index that already understood the code.
• JunieAI (JetBrains) the page returns 404
  JetBrains' agent that planned, wrote and tested inside their IDEs. That address now returns 404 and the feature sits inside the IDEs.
  Folded into the IDEs it lived in.
• Jupyter AI
  Model assistance inside JupyterLab, where the notebook cell is the unit rather than the file.
  Meets data work where it already happens.
• Kilo Code
  An open-source agent for VS Code, JetBrains and the CLI. Brings your own API keys at no markup and supports local models.
  Open source, no markup, keys stay yours.
• Kiro
  Turns prompts into executable specifications, then checks the code against them, rather than generating and hoping.
  Writes the spec first, which is the unusual part.
• Kodus
  Open-source code review that learns a team's conventions and reviews against them rather than against a generic ruleset.
  The open alternative to the review bots.
• Kombai
  Generates front-end code from designs, aiming at production markup rather than an approximation of the layout.
  Design to code, judged on the code.
• Lovable
  Builds and hosts applications from conversation, grown out of the gpt-engineer experiment.
  The research project that became a business.
• Mage
  Generates a full-stack React, Node and Prisma application on the Wasp framework from a description.
  Opinionated stack, which is why the output compiles.
• Magic
  A research company building models for very long context code work rather than a product you install today.
  A laboratory in a list of tools.
• Magic Patterns
  Turns prompts into interface code, constrained by an existing design system rather than inventing components.
  Prototypes in your own components, not generic ones.
• MagicLoops
  Combines model steps and real code into small automations, for people who will not write the whole script.
  Half prompt, half program, deliberately.
• Mastra
  A TypeScript agent framework with memory, tools, MCP and tracing built in rather than assembled.
  Agents for the TypeScript side of the house.
• MeterCall
  Runs agents and workflows locally with dry-run defaults and your own keys, so nothing leaves the machine.
  Local-first, and dry-run until you say otherwise.
• Microsoft Power Apps
  Low-code business applications inside the Microsoft estate, reaching the data those organisations already hold.
  Wins wherever the answer is already 'we use Microsoft'.
• Open Interpreter
  An agent that writes and runs code locally to complete a task, with the machine as the tool.
  Gave a model a shell, early and deliberately.
• OpenAI Codex CLI
  OpenAI's terminal coding agent, running against their models with local file and command access.
  OpenAI's answer in the terminal.
• OpenASE
  Turns tickets into working code, with agents picking up issues rather than being prompted.
  The queue is the interface.
• OpenCode
  An open-source terminal coding agent, model-agnostic, with the whole loop visible in the repository.
  The agent loop, readable.
• OpenHands
  The open-source, model-agnostic platform for cloud coding agents, formerly OpenDevin. The active project.
  The one that kept going after the rename.
• OpenHands (OpenDevin) the domain does not resolve
  The old OpenDevin address. The domain no longer resolves; the project continues as OpenHands.
  Renamed: the work moved to a new address.
• OpenMagic
  A toolbar over any running web app: select an element, describe a change, and it edits the real source.
  Points at the rendered page and edits the file behind it.
• PearAI the certificate does not validate
  An open-source editor fork with model features. The site's certificate has expired.
  Unreachable: the certificate expired.
• Phind the page returns 404
  A search and answer engine for developers. The site now returns 404.
  Gone: the page is not there.
• Plandex
  An open-source terminal agent built for large projects, with planning and change staging across many files.
  Designed for repositories too big for one context window.
• Potpie
  Builds task-specific agents against a knowledge graph of one codebase, rather than one general agent for everything.
  One agent per job, grounded in your repository.
• PraisonAI
  A framework for autonomous agents that research, plan and code, aimed at removing boilerplate work.
  Sells a workforce, not a library.
• Pythagora
  Listed as an automated development platform. The address now serves Pazi, a differently positioned product.
  Repositioned under a new name.
• Qoder
  Alibaba's programming platform with completion, conversational coding and generation.
  The Chinese market's entry, with global ambitions.
• Refact.ai
  An open-source agent for coding, debugging and testing, self-hostable with full repository context.
  Open, and runnable inside your own network.
• Replit
  A browser development environment that turned into an app builder, with hosting and databases attached.
  Started as an IDE for learners and became a factory.
• Replit Agent
  Replit's agent, which plans, builds and deploys inside the same environment it runs in.
  The agent and the runtime are the same product.
• Replit AI
  Replit's assistant features, now largely folded into the agent experience.
  Absorbed into the agent.
• Rocket.new the site refused an automated request
  Rapid application development from natural language. The site refused an automated request.
  Generation aimed at speed above all.
• Roo Code
  An open-source agent fork that grew its own direction. The address now presents Roomote, a hosted teammate that opens pull requests.
  The fork that became its own product.
• Rosebud AI
  Generates playable games from prompts, with deployment included.
  Games as the output, which is a harder target than a page.
• Rovo Dev (Atlassian)
  Atlassian's development agent, reaching into Jira and Confluence where the work is already described.
  Starts from the ticket, because Atlassian owns the ticket.
• Screenshot-to-Code
  Converts a screenshot into working HTML, React or Vue, one of the earliest demonstrations of the idea.
  The demo that convinced a lot of people at once.
• Smol Developer
  The first library to embed a developer agent inside another application, and an influential prototype.
  A small experiment that a lot of products grew out of.
• Softr
  Builds portals and internal tools over existing data sources without code.
  A front end for the spreadsheet that runs the business.
• StackSpot AI
  An enterprise development platform from Zup, built around reusable internal standards. Primarily documented in Portuguese.
  Built for one large organisation first, then sold.
• Stitch (Google)
  Google's interface generator for mobile and web, aimed at the early ideation stage.
  Built for the first draft, not the final screen.
• Supermaven
  Completion built for latency and a very large context window, competing on speed rather than on agent features.
  Chose to be fast rather than to be an agent.
• SwarmClaw
  A self-hosted runtime for multiple agents with memory, MCP tools, schedules and delegation between them.
  An agent runtime you can run on your own metal.
• SWE-agent
  Takes a GitHub issue and attempts a fix with the model of your choice. The academic work that grounded a lot of the benchmarks.
  The research the whole category cites.
• Tabby
  A self-contained, self-hosted completion server. No source leaves the network it runs on.
  Copilot's shape, inside your firewall.
• Tabnine
  A completion assistant that sells on where the code goes rather than on what the model can do: private deployment, no training on your source.
  Sold to the people who have to sign off, not the people who type.
• TeleportHQ
  A visual builder with headless content modelling and clean code export.
  Visual building that hands the code back.
• Tempo
  Turns feedback, issues and requirements into reviewed pull requests, with people approving the decisions that matter.
  Automates the work and keeps the judgement.
• tldraw Make Real
  Draw an interface on a canvas and get working code back. An early demonstration that shifted expectations.
  A sketch became a running page, and everyone noticed.
• Trae
  ByteDance's assistant and editor, positioned around collaboration with the developer rather than replacement.
  Another entry from outside the usual vendor set.
• Uizard
  Interface design for people without design training, working from templates and sketches.
  Design for the founder who has no designer.
• v0 by Vercel
  A duplicate entry for v0. This particular link has expired and no longer resolves to the product.
  Dead link: the same product lives at v0.dev.
• v0.dev
  Vercel's generative interface tool, producing React and Tailwind you can iterate on and then take into a project.
  Generates into the stack Vercel already owns.
• Visual Studio IntelliCode
  Microsoft's pre-Copilot completion work, ranking suggestions by what similar codebases actually did.
  Statistical completion, from before the models got large.
• Webflow AI
  Webflow's AI features for generating and optimising site content inside their visual builder.
  Generation inside a tool designers already use.
• Wiggum CLI
  An open autonomous agent that scans a stack, generates specifications through interviews, and works from them.
  Interviews you before it writes anything.
• Windsurf
  An agentic editor that was acquired and folded in. The address now serves Devin Desktop.
  Absorbed: the URL now belongs to somebody else.
• Zed
  A native, multiplayer editor from the people who built Atom and Tree-sitter, written for latency above all.
  The Atom team's second attempt, and much faster.

Other shelves