CursorAI code editor; Cursor 3 adds an agents window to run parallel agents.Claude CodeAnthropic's agentic coding tool: delegate engineering tasks from your terminal.LovableBuild and publish full-stack web apps by chatting with AI.Bolt.newPrompt, run, edit and deploy full-stack apps entirely in your browser.v0Vercel's prompt-to-app builder with Design Mode and one-click deploy.ReplitBrowser IDE whose Replit Agent builds and deploys apps from a description.Devin DesktopAgentic IDE with Cascade; renamed Devin Desktop by Cognition in June 2026.ZedOpen-source Rust editor at 120fps with an agent panel and parallel agents.GitHub CopilotCopilot agent mode in VS Code: edits files, runs commands, fixes its own errors.Codex CLIOpenAI's open-source coding agent that runs locally in your terminal.Gemini CLIGoogle's open-source terminal agent; free tier moved to Antigravity CLI in 2026.Base44Describe an app, get a working app with auth, database and hosting. Owned by Wix.n8nVisual workflow automation with AI agent nodes; self-host free.LangSmith Agent BuilderLangChain's no-code agent builder: describe the job, it writes its own instructions.DifyOpen-source platform for LLM apps: visual workflows, RAG and agents.RunwayGen-4.5 video generation with consistent characters and worlds.Kling AIKuaishou's video model: 15s clips, native audio, multi-shot storyboards.Google VeoGoogle DeepMind's video model with native sound, in Flow and the Gemini app.PikaPlayful text/image-to-video with effects like Pikaffects.HeyGenTurn one recording into a realistic AI avatar that presents any script.MidjourneyDistinctive art-quality images, now animated with V1 video.IdeogramImage generator famous for clean, accurate text in logos and posters.RecraftDesign-grade images and real vector SVGs, logos and brand styles.ClineOpen-source coding agent in VS Code with Plan/Act modes; bring any model.DevinCognition's autonomous AI software engineer, in the cloud or locally.KiroSpec-driven agentic IDE: turns prompts into specs, then code.Google AntigravityGoogle's agent-first IDE: agents plan, build and verify apps in a browser.WarpAgentic development environment: a modern terminal with built-in coding agents.Luma Dream MachineRay3 video model with HDR and fast Draft Mode.Adobe FireflyAdobe's commercially-safe image, video and vector generator.AiderOpen-source AI pair programming in your terminal, git-aware.JulesGoogle's asynchronous coding agent that works on your GitHub repos.SupabasePostgres backend with auth, storage and APIs; common backend for AI-built apps.CrewAIFramework for orchestrating role-playing, collaborative AI agents.LlamaIndexFramework for building agents over your data.Zapier AgentsBuild AI agents that act across Zapier's app integrations.OpenRouterOne API for many LLMs from different providers.OllamaRun open models locally on your machine.E2BSecure cloud sandboxes for running AI-generated code.BrowserbaseCloud headless browser infrastructure for AI agents.ScenarioAI generation of consistent game art and assets.MeshyGenerate 3D models from text or images.Rosebud AICreate browser games by describing them in plain language.ElevenLabsAI voice generation and sound effects, usable for game audio.Context7Feeds up-to-date, version-specific library docs and code examples into Claude Code, Cursor and other agents via an MCP server or CLI skill.DeepWikiAuto-generated, question-answerable wiki for any public GitHub repo, from Cognition (makers of Devin); also available as an MCP server.MintlifyDeveloper docs platform (git + MDX) with a built-in AI assistant, AI agents that update docs, llms.txt and an MCP server for your docs.CodeRabbitAI reviewer that comments on every pull request (GitHub, GitLab and more), plus CLI and IDE reviews.GreptileAI code reviewer that indexes your whole codebase and comments on pull requests with full-repo context.OpenCodeOpen-source AI coding agent for the terminal, desktop and IDE; bring any model (Claude, GPT, Gemini, local) or use its free models.LM StudioDesktop app to download and run open models locally (llama.cpp and MLX) with an OpenAI-compatible local server.MastraOpen-source TypeScript framework for AI agents: agents, workflows, memory, evals and tracing.ComfyUIOpen-source node-graph app for running image and video models locally (or on Comfy Cloud); new open models usually get day-one support.SeedanceByteDance's video generation model (Seedance 2.5), used directly and inside apps like CapCut, Dreamina, Picsart and others.Novastart OSAn infinite-canvas desktop OS where apps, files, shared canvases and AI agents live together; runs as an app on Windows/Intel Mac or as a full OS.Video UpscalerOnline AI video enhancer: upload a clip, upscale the resolution, reduce blur and noise, and download a sharper video, no install needed.TecAdRise BotA local Windows workspace for Cursor-powered agents, shared memory, scheduled routines and MCP tools. The app is free; Cursor API usage is billed separately.PixolmCreate campaign images and refine them through chat: remove objects, annotate changes and keep editing the same visual.OrkasDirect a team of AI agents by chat to research, write documents, build slides and create apps. Use managed models or connect your own provider.Monobot.AIBuild voice and chat agents with visual conversation flows, knowledge bases and business integrations for support and bookings.FreyaVideoAnimate product photos and create short videos or images with several models. Use one-time credit packs or an optional monthly plan.Magic HourCreate and edit videos, images and audio in a browser, with reusable character and object references plus API access.NewYouGoCreate images, videos and music in a browser. Signed-in users get 20 free FLUX.1 Schnell images daily; other models use paid credits. Standouts include Hotel Lobby AI, which turns two portraits into a 15-second duet performance video (700 credits per video), and Nano Banana 2.1 for text-to-image and image-to-image work.NEON Dance AITurn two adult photos or two pet photos into a 23-second duo dance video with music. Choose a main dancer and partner, then download the finished MP4.DocTreeTurn topics and source material into structured learning articles with citations and diagrams. Set your level and reading time, then ask an AI companion for help as you read.Learn AnythingUse an AI coding agent to turn a subject into an offline study app with guides, quizzes and timed exams. The MIT-licensed template is free; agent and model usage costs are separate.MangaTranslateTranslate manga images and PDFs with OCR, text replacement and layout editing. Choose source and target languages, then review or download the translated pages.VideoBGRemoverRemove the background from an existing video, preview the cutout and export transparent WebM or an MP4 with a new background. Includes REST API and MCP access.CileviRecord Windows demos with automatic zooms and a smooth cursor. Generate captions locally with Whisper, edit by deleting transcript words, and export MP4 or GIF.FORMWEFT Motion WorkbenchImport existing SRT subtitles and local video, edit animated caption cards, and export a finished video with the original audio. Basic editing works in your browser without an account.We0 AIBuild a website from a prompt, revise it through chat and test an interactive preview. Includes design, content-management and publishing tools, with one-time signup credits to try generation.OrbiTurn labelled GitHub issues into code changes, tests and pull requests. A separate AI session reviews and repairs the work before merge; release tasks can publish tags and notes.ForgeSpriteCreate static 2D game icons and sprites from a prompt. Review the plan and sample, edit assets, remove solid backgrounds and export PNGs or pack existing images into sprite sheets.VidmoatEdit existing footage through chat, review the changes on an editable timeline and export an MP4. Trim clips, add titles and revise the cut in your browser.HOL GuardLocal-first runtime control for AI coding agents. Runs beside Claude Code, Codex and Cursor and holds risky shell commands, secret reads, MCP calls and package installs for your approval.MomenticAI QA platform: write end-to-end tests in plain English that agents run and maintain, plus an AI agent that explores your app and reports bugs with repro steps.Playwright MCPMicrosoft's official MCP server that lets Claude Code, Cursor and other agents drive a real browser through Playwright to click through, test and debug web apps.ShipCheckPaste a public GitHub repo or live URL and get a launch report: leaked keys, committed .env files, missing security headers, broken links and slow pages, with copy-paste fix prompts.Ship SafeOpen-source CLI that scans your codebase, agents and MCP servers for LLM vulnerabilities, secrets, Supabase RLS misconfigurations and dependency CVEs, with CI checks and fix guidance.VibeSafe24 security checks for vibe-coded apps: a pre-launch scan of your repo (secrets, OWASP issues, Supabase RLS, Firebase rules) and a post-launch scan of the live site (SSL, headers, exposed env files).ConvexReactive backend where database, server functions, auth and realtime sync are all written in TypeScript, a format coding agents generate reliably.LaminarOpen-source platform to trace, evaluate and debug AI agents: see every step and LLM call, run evals and surface agent failures.OpikComet's open-source platform to trace, evaluate and monitor LLM apps and agents, with an agent playground and optimizer.RetainDBPersistent memory and context infrastructure for AI agents, with an SDK, MCP server and memory router so agents remember users and past work.memnodeHosted and local memory for AI agents with provenance, typed recall and garbage collection; usable through MCP locally or as a hosted tenant.GRIFFaiAgent OS that starts with governed shared memory for agents, then adds multi-agent coordination, human approvals, audit records and cost visibility.CoolifyOpen-source, self-hostable alternative to Heroku/Vercel: deploy apps, databases and 360+ services to your own server with git push deploys.BriefPatchTurn a client request into a structured project brief, keep a copy in your browser and export an HTML handoff. The core workspace is free without an account.Video Object RemoverRemove an unwanted object from existing footage with a text prompt. Review the selected mask, generate a preview and download the cleaned video.Zozo HQDispatch development issues to Claude Code, Codex or OpenCode, follow execution journals and move code through pull requests and checks. Runs locally or on a dedicated hosted stack.resolve() by WordLiftResolve names in text to Wikidata IDs or your own entity identifiers. Returns supported matches or an unresolved result for downstream agents and knowledge graphs.EtonoCreate and revise personal apps through conversation on an iPhone or Android phone. Run the resulting tools inside Etono and keep each project’s conversation and local data.Coding Tools MCPGive an MCP-compatible AI client workspace file reading, search, patches, command execution and Git inspection through a local coding-tool server.

Tool review standards

Version 1.6 · Effective · Change log

This page sets out the standards VibeCodePicks applies to every tool submitted for listing. The same standards apply to every submitter. When we make a decision on a submission, we refer to the relevant section of this page.

1. Scope

VibeCodePicks is a directory that helps users compare and choose AI tools by their goals. We consider submissions in the following areas:

Tools outside these areas are generally declined as out of scope (see reason 6.4).

2. Submission requirements

Submissions are made through the free submission form on our website (no account needed). Every submission goes into the same review queue and is reviewed the same way. We do not accept submissions by email. Choose the submission type that fits your tool: a commercial product or an open-source project.

2.1 Commercial products: what counts as a live product

A product counts as live only if it meets all four of the following:

  1. Open access: anyone can access and use it directly. A waitlist, invitation-only, pre-sale or "coming soon" page does not count; paid products that work right after payment are fine.
  2. Working core features: the core features genuinely work. A mock-up, a demo that only shows sample output, or fake data does not count.
  3. Stable official URL: the product has a stable official web address, not a temporary test or preview link.
  4. Maintained: the product has been updated or maintained within the last 90 days.

Required for commercial products:

  1. A live product as defined above.
  2. Official website. The product's own website, operated by the team that makes it.
  3. Demo video. Supplied by the team and meeting the demo video requirements (section 2.4). Any publicly reachable link works; it does not have to be on YouTube.
  4. Pricing or free-tier page. A public page that states what the product costs, or what can be used for free.
  5. Contact email. An address where we can send review notices.
  6. Optional: GitHub repository URL. Only if the tool has a public repository. Many tools do not, and this does not affect the review.

2.2 Open-source projects

We list good open-source projects, including early ones. Required:

  1. Installable: an installable release (for example a package, binary or extension) or clear install steps.
  2. Open-source license: a recognised open-source license in the repository.
  3. Active: commits within the last 6 months.
  4. README: a README that explains what the project does.
  5. Public repository URL, a demo video and a contact email. You supply your own demo video, meeting the demo video requirements (section 2.4). A video file in your repository or a GitHub release asset is fine.

The repository can serve as the official website, so a separate website or pricing page is not needed; just state any real costs to use the project (for example, a required paid API). Open-source projects use this 6-month activity window instead of the 90-day window in 2.1. Supporting evidence such as GitHub stars, discussion on social platforms, or being used or referenced by other projects counts in a project's favour, but there is no hard minimum (for example, no star threshold), so new projects are not excluded. The core features must still genuinely work (2.1, point 2).

If you have them, send at least 5 links from at least 2 different platforms to genuine reviews or posts by users who are not part of your team, or screenshots that include the original source (for example, the post URL and author). We gather the rest independently. Feedback without a verifiable source is not accepted. Feedback affects placement, not whether a live product is listed.

2.4 Demo video requirements (all submission types)

The same requirements apply to every submission type: commercial products, open-source projects and Arena entries that include a video. The video must meet these requirements, and only these:

  1. Shows what the product does.
  2. Is clear (readable and in focus).
  3. Is at least 3 seconds long.
  4. Leaks no private data (for example personal details, API keys or customer data).
  5. Has no malicious or misleading calls to action.

It does not have to be on YouTube. Any publicly reachable video link is accepted: a direct .mp4 or .webm file (a GitHub release asset or a file in your repository is fine), Vimeo or YouTube. We embed the video directly on our site; for video files not on YouTube we may host a copy ourselves so the page stays fast and the video stays available.

No video yet? Video production by VibeCodePicks is available as a paid service. Our videos follow the style of our product-intro screencasts (see the templates below): no real person on camera, clean UI animation. You can also record the video yourself; any recording that meets the requirements above is accepted.

Video production prices (USD, one-off):

Every package supports Template A, B or C. Packages set the length, included work, revisions and delivery time; templates set the visual style. Choose them separately on the submission form.

If the product has many features, you decide which features or workflows to highlight. We ask for reasonable materials: the product link, a test account or access if needed, the 1–3 key features to show, logo and brand colours, and your own recording if you choose Lite. For a video that includes a hands-on review, we can try the product and record it ourselves. The delivery clock starts once the materials are complete. Ordering or paying for video production never affects listing or review decisions. Ordering opens soon — there is no payment link yet; tick the video production option on the submission form and we will contact you.

Demo video templates (our production style)

All three styles are available with every package: Lite US$29, Standard US$69 and Plus US$129. Choose your package for the length, included work, revisions and delivery time; choose A, B or C for the visual style. These clips are style previews, not package examples. See prices and materials, then select your package and style on the submission form. Ordering opens soon; we will contact you. You can also submit your own video. Ordering never affects listing.

Template A: Phone assistant, split-screen
Vertical. The AI chat on one side, the app on the other, with dashed lines linking each message to the automated taps it triggers.
Template B: Illustrated product intro
Horizontal. A warm, hand-drawn notebook style that explains what the product does and who it is for.
Template C: Architecture explainer
Horizontal. Technical diagrams and code that show how the product works step by step.

3. How we verify

ItemWhat we check
Live productWe open the product and check all four points in 2.1: a user can start without an invitation, waitlist or pre-sale (paying and using it right away is fine); the core features work with real input; the URL is the stable official one; and there is a visible update (changelog, release, blog or commit) within the last 90 days.
Official websiteThe site loads, belongs to the product, and is consistent with the submitted name and description.
Open-source repositoryThe license file, the README, the install steps or latest release, and the date of the latest commit (within 6 months). We note stars, discussions and dependents as supporting evidence only.
Demo videoThe link is publicly reachable and the video meets the five requirements in 2.4: it shows what the product does, is clear, is at least 3 seconds long, leaks no private data and has no malicious or misleading calls to action.
PricingThe pricing or free-tier page loads and matches what the product offers. Not needed for open-source projects.
Contact emailUsed to send the acknowledgement and review notices.
Independent user feedbackEach source is open and public, is written by someone outside the submitting team, and refers to this product. We count it against the rules in section 4. Screenshots must show where the original can be found.

If an item is missing or cannot be checked, we first ask for it (see timeline) rather than declining.

4. Placement and user feedback

Rankings are never sold. When two tools match equally well, the one with an exclusive VCP user coupon is marked.

User feedback threshold for comparison pages

A handful of reviews cannot represent how a whole market sees a tool, so comparison pages use a published minimum:

  1. At least 10 independent user reviews or posts about the tool,
  2. from at least 2 different platforms (for example X, Reddit, Hacker News, Product Hunt, YouTube or blogs),
  3. all published within the last 6 months.

How we count:

The threshold is about how much independent evidence exists, not about whether the evidence is positive. Positive, mixed and negative posts all count, and we show them as they are. For transparency, a tool's community feedback is published with its data: the number of posts sampled (as in "Based on N public posts" and "N mentions" on our pages), the platforms they come from, the date range, and the positive, mixed and negative split, counted per point rather than per post (one post can back several points). Each tool page has a "Community feedback data" table per use case: mentions, then the posts behind its recurring praise (positive), disputed points marked "Split" (mixed) and criticism (negative). We show only what we sampled; a tool with no sampled posts says so.

Order within pages follows our editorial method, which is based on public community sentiment and the user's stated goal, and is the same for submitted and non-submitted tools.

5. Labels

6. Reasons we may decline

A declined submission receives one of the following reasons, which links to the matching item below.

  1. Not a live product (not-public). The product does not yet meet our definition of a live product (section 2.1): for example, waitlist or pre-sale only, or core features are only a demo or mock-up.
  2. Information could not be verified (unverifiable). The submitted information could not be verified.
  3. Outside our scope (out-of-scope). The product is outside the scope of our directory.
  4. Already listed (duplicate). The tool is already listed on VibeCodePicks.

We may also decline submissions that are spam, misleading, unlawful, or that copy another product without meaningful differences; these are recorded as "information could not be verified" or "outside our scope", as applicable.

7. What never influences listing or ranking

8. Review timeline, resubmission and appeals

  1. Acknowledgement: sent immediately after submission, as a comment on the submission and by email.
  2. Review: submissions are reviewed in the order received. We aim to complete a review within 7 days; this is a target, not a guarantee.
  3. Result: one of: listed (homepage and comparison pages, or homepage only), more information needed, or not listed. Each result is posted on the submission and emailed to the contact address.
  4. More information needed: the notice lists exactly what is missing. Reply to the notice email (or on the submission) to add it, and the review resumes. If there is no reply within 14 days, the submission is closed for now; this is not a final decision.
  5. Resubmission: a closed or declined submission can be submitted again at any time through the submission page once the issue has been resolved.
  6. Appeals: if you believe a decision was made in error, reply to the notice email (or on the submission) and we will review it again.

9. Updates, corrections and removal

10. How submission data is handled

11. Change log

VersionEffectiveChanges
1.6Clarifications: paid products that work right after payment qualify; open-source projects use the 6-month window and state any real costs; feedback split counted per point; paid videos and Arena entries are not independent feedback.
1.5Published paid video production prices in section 2.4 (Lite US$29 / Standard US$69 / Plus US$129), delivery times, materials we ask for, and the rule that ordering never affects listing; ordering opens soon via the form checkbox (no payment link yet).
1.4Described where the published feedback data appears (the "Community feedback data" table on each tool page) and how positive, mixed and negative are counted; added our demo video templates (A, B, C) to section 2.4 for the paid video-production service; trimmed the list of example video hosts (any publicly reachable link is still accepted). No change to the rules.
1.3Added demo video requirements for all submission types (section 2.4): shows what the product does, clear, at least 3 seconds, no private data, no malicious or misleading calls to action; YouTube no longer preferred, any publicly reachable video link accepted (including direct .mp4/.webm files, Vimeo); we embed it and may host a copy; video production by VibeCodePicks offered as a paid service (pricing coming soon).
1.2Defined a live product (four conditions: open access, working core features, stable official URL, maintained within 90 days); added requirements for open-source projects (no star minimum; repository can serve as the website, open source counts as free pricing); replaced the "2–3 links" feedback rule with a published threshold for comparison pages (at least 10 independent reviews from at least 2 platforms within 6 months, at most 2 per account, vendor links at most half, 90-day resampling) and published feedback data; decline reason 6.1 now refers to the live-product definition; linked the separate Open Call & Arena rules.
1.1Added the submission form on our website (no account needed) as the main channel, with the contact email kept private; GitHub repository link added as an optional field; email submissions are no longer accepted; spam protection described.
1.0First published version.

Changes to these standards are recorded here with a new version number. A submission is reviewed under the version in effect when the review is made.

Submit a tool for free →

See also: Submit page · Open Call & Arena · About · Privacy · Contact