Most of what I cover on this site is AI for everyday people. This page is the exception: it’s my running list of new AI dev tools — developer-focused AI launches I’ve spotted, mostly on Product Hunt, that looked worth knowing about. I want to be upfront: I haven’t tested the tools on this page. I don’t test developer tools, but it’s a huge part of the AI world, and I’d rather bring these launches to developers’ attention than pretend the category doesn’t exist. What you’ll find here, and in each linked guide, is information pulled from the vendors’ own sites and documentation, plus any early community feedback I could find — organised so you can decide quickly whether a tool deserves a spot in your stack. Where I have properly tested something, it gets a full scored review instead, and I’ll say so.
One more thing: comments are open on every guide. If you use one of these tools, leave a comment on its guide saying how you find it — what works, what doesn’t, what you’d pick instead. You know this territory better than I do, and honest developer experience helps everyone who lands here after you.
Which AI Dev Tools Have a Full Write-Up?
These tools have their own guide on WhatAITools, covering what each one does, who it’s for and what it costs. The newest are at the top. New guides are added here as they publish.
Base Code by Base44 — non-engineers propose changes to an existing GitHub repo in plain language, landing as ordinary pull requests
EmbeddingGemma 2 — Google’s open, lightweight multimodal embedding model for local search and RAG
Manifold MCP — SEO, AI-answers, social, ads and leads data served to your agent over MCP
Cube Forms — forms designed to be filled in by AI agents as well as humans, over MCP or plain HTTP
Scrape Anything by Context.dev — turn any website into clean, structured data with one API call
ExcaliClaw — hand-drawn-style diagrams your AI agent can draw with you over MCP
guilds.run — self-hosted multi-agent workflows you can fully inspect: every prompt, tool call and memory change
Codemind — delegate coding tasks to a build service and pay per verified build; failed builds cost nothing
Dropstone CLI — a self-learning coding agent for your terminal, running the vendor’s own Fast/Pro/Heavy models
Which New AI Dev Tools Are on My Radar?
The newest launches are at the top. All pricing below comes from each vendor’s own pages on the date shown for each batch — check before you commit, because launch pricing changes fast. Guides for each tool are published as they’re ready and linked from the tables.
Added 23 October 2026
| Tool | What it does | Pricing (vendor, 23 Oct 2026) |
|---|---|---|
| heal.dev | Autonomous agent that runs your app in a production-like sandbox and only reports bugs it reproduces with a failing test | Free during beta, no card; free for open source; no prices published |
| CleanSlop | Scans vibe-coded TypeScript apps for what breaks under real users — findings ranked, agent-fixable | Free 1 scan/mo; plans $19–$149/mo; credit packs $20–$110 never expire (currency unnamed) |
| Whiteboard by dev.fast | Open-source IDE where humans and agents design software on a shared canvas, with diagrams linked to code | Free, MIT; hosted team product planned, unpriced |
| Tetriz Engineering Intelligence | Open-source Go packages linking AI coding sessions to the PRs they produced, with complexity and ROI analysis | Free, MIT — a library you wire up, not a dashboard |
| Mercury 3 | Inception’s diffusion LLM generating tokens in parallel — vendor claims 10x speed and a 1M context | $0.10/1M input, $0.35/1M output tokens (USD), marked 50% off with no end date |
| SuperCMO AI | Open-source skills that turn Claude Code, Cursor or Codex into an AI marketing lead | Free (Apache-2.0) on your own keys; pay-per-use credits unpriced |
| Bits | Open-source, local-first, always-on personal agent with plain-Markdown memory | Free, MIT — bring your own model keys or run local models |
Added 22 October 2026
| Tool | What it does | Pricing (vendor, 22 Oct 2026) |
|---|---|---|
| BAND for Developers | Shared rooms where coding agents (Claude Code, Codex, Copilot, OpenCode) and humans coordinate, with governance and audit built in | Desktop app free; enterprise platform unpriced |
| Mouse-1 | Fast decision model for agent harnesses — returns probabilities over fixed options, so it can’t be prompt-injected off them (vendor claim) | $0.10 per 1,000 decisions or $0.10 per 1M input tokens (USD); no documented free tier |
| Triage by Chodeberg | Replays user-reported bugs in clean sandboxes — only reports that reproduce reach you, with recording, backtraces and a README | 100 free credits/mo; 1¢ per minute after; packs $100–$1,000, credits never expire (USD) |
| ilha | Mac notch app tracking Claude Code, Codex and Grok Build limits, with one-click and automatic account switching | Free tier; Pro $20 one-time per Mac; 14-day trial, no card (USD) |
| Agent Pets | Your coding agent sessions as pets walking along every Chrome page — approve, deny and reply from any tab, all local | $6 one-time launch price (regular $25, USD); 7-day refund |
| Nimbalyst Wiki | A wiki your agents write and maintain — decisions, open questions and status as plain markdown in your repo | Free, MIT open source; hosted Teams add-on unpriced |
Added 21 October 2026
| Tool | What it does | Pricing (vendor, 21 Oct 2026) |
|---|---|---|
| Harness² | Google Research framework letting an LLM agent optimise its own prompts, tools and memory per task — no retraining | Free, open source (GitHub) |
| RunInfra | Hosted open models (GLM, DeepSeek, Qwen) behind one key that works with Claude Code, Codex and other agents | Starter $10/mo, Pro $29/mo, Team $99/mo (USD); included usage not quantified |
| Praex | Coding agent running its own Velox/Faber/Lucia models; MIT fork of opencode; terminal, web and phone | Free tier; Pro $20/mo (USD); free on your own keys or hardware |
| call4me | Gives your agent a real phone over MCP — calls businesses, navigates menus, waits on hold, reports back with transcript | Prepaid credits from $10; $0.25/min from pickup, hold half price (USD) |
| yggy | Self-building knowledge graph from docs, mail and tickets — source-linked facts, queryable via API, CLI and MCP; EU-hosted | Free to start per PH listing; usage-based billing, exact figures not published readably |
Added 20 October 2026
| Tool | What it does | Pricing (vendor, 20 Oct 2026) |
|---|---|---|
| shim-cli | Open-source hook/proxy that shows and masks what your coding agent sends to the model | Free, open source (Apache-2.0) |
| CrawlForge MCP | 31 web tools (scrape, search, crawl, extract) for AI agents in one open-source MCP server | 1,000 free credits, then pay-per-use from 1 credit/call |
| Starkomand | Runs your AI coding agents side by side on an infinite canvas on your Mac | “No account, no subscription” — no pricing or licence published |
| Lancetta | Mac menu-bar app that projects your Claude Code/Codex quota and says when to push or ease off | Free; licence not stated |
| buddylist | Your coding agents as an MSN-style buddy list — solo worktrees or group chats on one project | Free (optional Buy Me a Coffee) |
| walkai-talkai | Talk to your coding agent over a real handheld radio; local Whisper transcription | One-time lifetime licence (price unlisted; 50% launch discount shown) |
| Beam.win | Opens individual windows from a headless Mac running your agents, as if local | Free to try 20 min/day; full pricing unpublished |
| Point and Yap | Point at your screen and talk — both go to your coding agent via MCP | Free, open source |
| Framefields | Code-first video engine on native WebGPU in Node.js — no headless browser | Free, open source |
| 3AM Super Intelligence | AI on-call engineer: investigates incidents, proves root cause, fixes after approval | Open alpha; first 90 days free, no standard pricing yet |
| Popcraft | MCP server letting agents generate video, images, music, SFX and 3D with cost previews | 100 free credits; subscription + credits, details unpublished |
| Beam (Reflection AI) | 501B-parameter open-weights MoE model for coding and agentic work, 1M context | Weights free under Apache 2.0 — you pay for the hardware |
| AdHelm | Open-source CLI connecting Claude Code/Codex to your ad accounts to audit and recommend | Free, open source (Apache-2.0) |
Added 13 October 2026
| Tool | What it does | Pricing (vendor, 13 Oct 2026) |
|---|---|---|
| Dropstone CLI | A self-learning coding agent for your terminal, running the vendor’s own Fast/Pro/Heavy models | Free tier; Lite $5/mo, Pro $15/mo, Max from $75/mo (USD) |
| Codemind | Delegate coding tasks and pay per verified build — failed builds cost nothing | 100 free verified builds first month, then 50/mo; $0.10 per verified build (top-ups not yet live) |
| guilds.run | Build multi-agent workflows with checks you can trust, self-hosted via Docker | Free to self-host (bring your own model keys); hosted version coming soon, unpriced |
| ExcaliClaw | Hand-drawn-style diagrams your AI agent can draw with you, via MCP | Free plan; Starter €9/mo; Pro €29/mo (EUR) |
| Scrape Anything by Context.dev | Turn any website into clean, structured data with one API call | Free 1,000 credits/mo; paid from $19/mo to $499/mo |
| Cube Forms | Forms designed to be filled in by AI agents as well as humans | Free — no paid plans published yet |
| Manifold MCP | SEO, AI-answers, social, ads and leads data served to your agent over MCP | Free starter credits; Pro from $20/mo for 2,500 credits (USD) |
| EmbeddingGemma 2 | Google’s open, lightweight multimodal embedding model (text, vision, audio) | Weights free under Apache 2.0 — you pay for your own hardware |
| Base Code by Base44 | Lets your whole team build on your product inside the Base44 platform | No separate price published; Base44 plans run free to $160/mo (USD) |
How Do I Choose Which AI Dev Tools to Feature?
Each tool on this page gets its own information guide covering what it is, what it costs, who it’s actually for, and — where anyone is saying anything, because most of these are days old — what developers make of it. I don’t score tools I haven’t used, I don’t take payment for a listing, and I note on every guide whether the vendor runs an affiliate programme. If a tool here turns out to be something I can meaningfully test, it graduates to a full review with a score.
If you’re not a developer and landed here by accident — the rest of the site is for you: plain-English, hands-on reviews of AI tools anyone can use.