Today's drop: 3 new tools, 1 listing update
The directory now lists 109 hand-checked tools. New today: Recursive Flint, an Apache-2.0 coding model that runs on a laptop; OpenAI's Decisions API, now generally available for fast yes/no, pick-one and score calls; and PentAGI, a self-hosted AI pentesting system with about 32,000 GitHub stars. AI for IT & Cybersecurity grows to 31 tools. Pricing only where the official page states it.
Previous drop: Oct 22, 2026 →
Get the weekly dropThe best new AI tools, one email a week. Free.
New in the directory 3
Browse all 109Tool
Category
Price
Status
Link
Recursive FlintRecursive's first model: an open-weight mixture-of-experts coding model (39.8B total, about 2.5B active per token) built for fast agentic coding on your own machine. Recursive says it leads MoE models under 50B parameters on SWE-Bench Pro v2, Terminal-Bench 2.1 and SWE-Bench Verified, and runs at 125+ tokens/sec on an M4 Pro MacBook Pro. Runs via Ollama (flint:40b), llama.cpp, LM Studio, Unsloth, vLLM or SGLang; the INT4 GGUF needs a 48GB Apple Silicon Mac and Unsloth's 3-bit GGUF fits 24GB RAM. Released Oct 21, 2026; Recursive says it may need more step-by-step guidance than big closed models.
Price Free open weights (Apache-2.0)
OpenAI Decisions APIOpenAI API for the small, frequent decisions inside apps and agents: classify an input, pick a route or choose the next action. You define a question and the allowed answers, send text, images or JSON, and get back a probability, a choice or a score instead of free text. Built on GPT-6 Luna; OpenAI says it's up to 10x faster than Luna through the Responses API. Generally available Oct 22, 2026, with up to 400 questions per request, prompt caching and GPT-Live integration.
Price $0.10 per 1M input tokens ($0.01 cached); no output-token charges
PentAGIOpen-source, self-hosted multi-agent system for autonomous penetration testing from VXControl. Describe a target and goal in plain language and a team of AI agents plans the work, runs 20+ tools such as nmap, Metasploit and sqlmap inside an isolated Docker sandbox, remembers what worked and writes a report, with a web UI plus REST and GraphQL APIs. Works with 10+ cloud model providers or local models through Ollama or vLLM. About 32,000 GitHub stars; v2.2.0 shipped Oct 5, 2026. For authorized testing only.
Price Open source (MIT); PentAGI Enterprise available
Why these three
- Recursive Flint is Recursive's first model: 39.8B parameters in total but only about 2.5B active per token, so it runs fast on local hardware. Recursive reports 125+ tokens a second on an M4 Pro MacBook Pro and says it leads mixture-of-experts models under 50B parameters on SWE-Bench Pro v2, Terminal-Bench 2.1 and SWE-Bench Verified (vendor benchmarks). The weights are Apache-2.0 on Hugging Face and it's already in Ollama as
flint:40b. Hardware reality: the official INT4 GGUF needs a 48GB Apple Silicon Mac, and Unsloth's 3-bit GGUF fits 24GB of RAM at short contexts. Recursive itself says to break complex work into smaller steps. - OpenAI Decisions API went GA on Oct 22. Instead of generating text, it answers a question you define with a probability, one of your allowed choices, or a score, from text, images or JSON. It runs on GPT-6 Luna, costs $0.10 per million input tokens ($0.01 cached) with no output-token charges, and OpenAI says it's up to 10x faster than asking Luna through the Responses API. Good fit for routing, triage, guardrails and agent "should I continue?" checks. Datadog told OpenAI it uses it to score the risk of agent-generated shell commands before running them.
- PentAGI from VXControl is a self-hosted, MIT-licensed multi-agent pentesting system: describe a target and goal, and its agents plan the work, run 20+ tools such as nmap, Metasploit and sqlmap in an isolated Docker sandbox and write a report. Unlike Strix and Shannon, which focus on your app's code, PentAGI works more like a full network pentest and can run on local models through Ollama or vLLM. About 32,000 GitHub stars; v2.2.0 shipped Oct 5. Authorized targets only.
Listing updates
Checked on Oct 23, 2026 against official pages and announcements.
- Ollama: Recursive Flint is in the Ollama library (
ollama run flint:40b), so you can try it without building llama.cpp yourself.
Category changes
- AI for IT & Cybersecurity grows from 30 to 31 tools. The Breaking into IT guide adds PentAGI to the vulnerability scanning step as a self-hosted lab option that works with local models.
- Coding adds Recursive Flint next to NVIDIA Nemotron 3.5 Bolt as a second open local coding model, and Agents & Automation adds the OpenAI Decisions API.
Checked, not listed
- Mica v0.5: an Apache-2.0 4B "decision model" that returns probabilities for yes/no, choice and score questions on one consumer GPU, an open-source cousin of the Decisions API idea. Interesting, but it's a one-person project with about 40 GitHub stars and its benchmarks are the author's own runs, so we'll watch it.