
AI tools for traders and creators, tested on real tasks. What it does, what it costs, what it gets wrong. Built by @enhale11. https://nitter.cf/t.co/OG29rinKRQ
Florida, United States
Joined January 2022
- Tweets11.7K
- Following1.7K
- Followers1.4K
- Likes10.7K
Pinned Tweet
This account is back with a new focus: AI tools for traders and creators. Each post tests one took on one real task. What it did, how long it took, what it got wrong.
Tool directory: https:// workswithai.pages.dev
Markets and AI for business: Enhale11
277
NVIDIA open-sourced SkillSpector, a scanner for agent and MCP skills.
Its README: of 31,132 skills studied, 26.1% had vulnerabilities and 5.2% looked malicious.
Free. The catch: --no-llm is static checks only; the deeper pass needs your own model.
github.com/NVIDIA/SkillSpect…
5
New external sandboxes from Simon Willison and Prime Radiant keep Claude Code, Codex or Pi on your Mac but run their commands in Linux, with only the folders and hosts you allow.
The gap: Claude Code sends only Bash. Its Read and Edit tools still work on your Mac directly.
1
9
Taan v1-preview scores music to a video's cuts, and you can change one instrument or one timing without regenerating the whole track.
Creator is $15 a month for 25 minutes of music, Agency $45 for 60.
The catch: orchestral and cinematic only for now.
taan.ai
9
ci-proto, a new open-weight time-series model, installs with pip, runs on CPU or GPU and returns nine quantiles instead of one line.
The catch for traders: a benchmark lead isn't an edge. Backtest it on your own market's data first.
pypi.org/project/continuousi…
8
Meta's Muse Ember lists at $0.10 per million input tokens and $0.50 output, with a 1M-token context window.
The catch is the 200K line. Go over it and the whole request bills at $0.30 and $1.50, cached tokens and carried-over conversation included.
6
Google's new Gemini Decision 1 doesn't write answers. It scores fixed options: yes/no, multiple choice or a rubric, up to 1,000 questions on one input.
$0.10 per million input tokens, nothing billed for output.
The catch: no Batch API and no context caching.
5
Gemini 4 Carbon is on the paid Gemini API at $2 per million input tokens and $10 per million output, through Nov 30.
On Dec 1 that doubles to $4 and $20.
The catch: thinking tokens bill as output, so price per token isn't price per finished task.
aistudio.google.com/docs/pri…
7
For podcast clips, who said it matters as much as what was said.
AirCaps A5S Async: $0.18 per audio hour, with speaker labels, word timestamps and SRT/VTT subtitles included. 5 free hours per account.
Limits: 3 hours per file, 9 languages, key-term boosting in English only.
1
1
9
Claude Sheets went into beta today on all paid Claude plans. It builds spreadsheets with tabs, formulas and dropdowns, and outlines each edit it makes.
For a trade log, that edit trail is the useful part.
The catch: Anthropic calls it simple spreadsheet functionality for now.
1
6
Adobe previewed a local MCP for Creative Cloud today. Your own agent can work in Photoshop, Illustrator, Premiere and After Effects on native files, with layers and sequences still editable.
The catch: it's not out yet. Private beta starts early 2027.
blog.adobe.com/en/publish/20…
21
kepano's obsidian-skills teaches Claude Code, Codex and OpenCode to write real Obsidian files: wikilinks, Bases, JSON Canvas, plus the Obsidian CLI.
Free, MIT.
The gap: it knows Obsidian's formats, not your vault's folder and naming rules. You still write those down.
15
Claude Code's "Yes, and don't ask again" saves the approved command in .claude/settings.local.json. If that command had a password in it, so does the file.
Apiiro found 1,122 public GitHub copies with secrets.
Check: git ls-files .claude
Rotate first, then untrack it.
11
Vercel's agent-browser lets an AI agent drive a browser from the shell. Each page element gets a ref like @e1, and the agent clicks the ref.
Free, Apache-2.0.
The catch: refs go stale after any page change, so the agent must take a new snapshot.
github.com/vercel-labs/agent…
8
Perplexity Teammates are agents that keep working on a standing job, like tracking competitors or drafting support replies.
The catch is access: it's a research preview, web only, and it needs the $200 a month Max plan. Enterprise accounts aren't included by default.
1
7
Glasshand rents AI agents a real iPhone for $60 a month or a Mac mini for $90, billed monthly with no hourly option.
The catch: a human still approves and pays at checkout, and each device is its own subscription.
21
Sand.ai's MAGI-2 Preview makes video with matching sound from a text prompt or a still image. Weights are open under Apache 2.0.
The catch for creators: clips are 10 seconds only, and the README calls for 8 NVIDIA Hopper GPUs.
github.com/SandAI-org/MAGI-2…
11
GitHub now lets you manage Copilot code review settings through the REST and GraphQL APIs, from enterprise policy down to a single repo.
Pro through Enterprise. Review overviews are more compact, and agent sessions show AI credits used per review.
github.blog/changelog/2026-1…
1
13
Google Jules stops new sessions on Nov 11, then goes read-only until it winds down Dec 31.
Export settings, memories, and session history with the Jules CLI before then. Your GitHub branches and PRs stay put.
jules.google/docs/changelog/…
18
Qoder's Qoduck stealth model is free daily for individual users from Nov 7.
Within the allowance it bills at 0.0x Credits (was 0.5x), even at max thinking. Teams/Enterprise and the IDE, JetBrains plugin, and CLI aren't covered.
docs.qoder.com/events/qoduck…
1
15
DuckDB's Claude Code plugin routes data jobs through the DuckDB CLI instead of guessed Python scripts.
Nine skills cover read, query, convert, S3 explore, and spatial. Install: /plugin install duckdb-skills@claude-plugins-official
duckdb.org/2026/09/16/duckdb…
1
12