AI Radar

A personally curated list of the best AI tools in their category at a given point in time — not a directory of everything that exists.

The focus is on niche and specific domains (3D, audio, video, TTS, OCR) that are rarely covered by mainstream AI tool aggregators, and on tools that have been personally tested or closely evaluated. Entries are dated so you can track how the landscape has evolved since 2024.

What makes this different from other lists:

Most AI tool directories (awesome-ai-tools, futuretools.io, theresanaiforthat.com) optimize for breadth — thousands of entries, shallow descriptions, no curation philosophy. This list goes the other direction:

• Best-of-class picks per category, not exhaustive catalogues
• Covers niche domains aggregators skip: 3D, audio synthesis, OCR, cybersecurity
• Includes academic research tools and papers alongside commercial products
• Explicit local/offline tagging — useful if you care about privacy or running without a cloud dependency
• Dated entries since 2024 — older entries show what was state-of-the-art at the time

---

Categories

Category	Description
AI Agents	Agentic coding apps, agent frameworks, multi-step AI automation
Video Generation	Text-to-video, image-to-video, live generation, world generation
Video Editing	Editing and transforming existing footage
Image Generation	Text-to-image models, platforms, research demos
Image Editing & Design	Image editing tools, upscalers, UI/design platforms
Voice, TTS & Transcription	Text-to-speech, voice cloning, speech-to-text, translation
Music & Audio	Music generation, AI singing, video-to-audio
3D Tools	Scene generation, asset creation, reconstruction, rigging
AI Models & LLMs	Niche and notable LLMs, local model runners, benchmark resources
General Tools	Dev tools, production platforms, search, browser extensions
Fine-Tuning & LoRA	Fine-tuning frameworks, model customization platforms
OCR & Document Understanding	Optical character recognition and document extraction
Cybersecurity	AI-powered security research and vulnerability tools
Web Scraping	Web data extraction and LLM-ready conversion tools

---

Entry Format

Every entry follows this structure:

#### [Tool Name](link) — `MM.YYYY` — `tag1` `tag2`
One-sentence description: what it does and what makes it notable.

Tags (use only the relevant ones):

Tag	Meaning
open-source	Weights or code are publicly available
local	Can run on local hardware (no cloud required)
online	Cloud-only, requires internet access
api	Primarily accessed via API
free	Completely free to use
freemium	Free tier available, paid tiers exist
paid	Paid only
research	Academic / non-production tool
platform	A platform hosting multiple models or tools

Entries missing a link, a confirmed date, or a specific tag simply omit that piece rather than being flagged with a placeholder comment.

---

Sources

Scouting is based on personal research and the following channels:

• The AI Search — news and tool discovery focused on newly released models and demos
• MrEflow — practical AI breakdowns with a hands-on workflow perspective
• The AI Search (Substack) and FutureTools News — scouted automatically via the /scout command, which surfaces candidates into new inputs.md for review

See sources.md for the full landscape comparison.

---

Scouting Workflow

Run /scout (a Claude Code project command, defined in .claude/commands/scout.md) whenever you want fresh candidates instead of watching videos:

1. It fetches new posts from The AI Search Substack (via RSS) and new items from FutureTools News, going back only as far as the last run.
2. It checks each candidate against every file in Categories/ and drops anything already in the list (including obvious version bumps of an already-catalogued tool).
3. Survivors get appended to new inputs.md, in the same raw one-line format used for manual entries, under a ## Scouted <date> heading.
4. It does not touch any Categories/*.md file itself — it only stages candidates for you to review.

Your side of the workflow stays the same as always:

1. Prune new inputs.md down to what's actually worth adding — delete the rest.
2. Ask Claude to merge what's left into the right Categories/*.md files (correct format, dedup/supersession per CONTRIBUTING.md, dates verified).
3. new inputs.md gets cleared back to just its header once merged, ready for the next batch.

/scout keeps its own state in .claude/scout-state.json (gitignored) so re-runs don't resurface things you already pruned out. Delete that file if you ever want a full re-scan.

---

Roadmap

• Automated scouting — done via the /scout command (see above); on-demand for now, could move to a scheduled run later
• Human-in-the-loop validation — every new entry reviewed before it lands in the list
• Structured comparisons — head-to-head notes within categories (e.g. best local video model in June 2025)

---

Contributing

See CONTRIBUTING.md.