Name: aLLMost — AI Prompt History & Response Analyzer
Author: algomittens

History

3 snapshots

Tracking since May 9, 2026.

Not enough history yet for this metric — the chart fills in as we collect more snapshots.

View as table

Date	Users	Rating	Reviews	Version
May 9, 2026	—	—	—	0.2.2
May 19, 2026	—	—	—	0.2.2
May 25, 2026	—	—	—	0.2.4
Now	1	—	—	0.2.4

Changelog

May 19, 2026

description

aLLMost is a thin local-first lens that overlays intelligence on the LLM chat sites you already use. No accounts. No telemetry. No backend.

It catches the moments where the model hedges, repeats itself, or confidently overclaims — and shows you, inline, what just happened.

HOW IT WORKS — three layers, one lens

▸ BEFORE the prompt — semantic match against your past prompts; intent-based model routing
▸ DURING the response — sentence-level confidence heatmap painted in place
▸ ACROSS conversations — quality scores per platform, reuse clusters, exportable history

THE 8 FEATURES (3 free + 5 Pro)

FREE — works forever, on every supported platform

▸ Déjà Prompt
As you type, aLLMost searches your captured prompt history (TF-IDF locally) and floats a "you've asked this before" card with the original conversation link. Stops you re-deriving answers you've already paid for.

▸ Confidence Heatmap
Every response sentence is scored against 300+ hedging, speculation, refusal, and overclaim patterns. Tints applied in place — amber (hedged), red (speculative), violet (overclaim). Toggle with Alt+H. Runs in under 20ms locally.

▸ Evasion Detection
Per-response evasion percentage derived from the same heatmap classifier. One-glance pill anchored bottom-right of every reply.

PRO — $7/month or $59/year (save 30%)

▸ Model Router
Classifies your prompt's intent (8 categories: coding, summarization, creative, math, etc.), looks up which model meaningfully outranks the others on public benchmarks for that task, and nudges you only when the gap is real. Never naggy.

▸ Quality Scores
Rolling 30-day quality score per platform, computed from your captures. Aggregates heatmap redness, instruction-follow signals, and your thumbs-up/down. Surfaces in the popup as a per-platform bar chart.

▸ Reuse Clusters
Online agglomerative clustering on your captured prompts (cosine similarity over TF vectors). Surfaces the topics you keep cycling through — "14 variations on React state in 3 platforms" — so you can stop having the same conversation again.

▸ Personalized Routing
The router starts from public benchmarks. Every accept, dismiss, and thumbs-up/down nudges your personal score per (intent, platform). Over time it diverges from the crowd toward what works for you.

▸ Prompts Export
Full local prompt history downloadable as a single JSON file — platform, prompt, evasion %, timestamp. Your data, in your hands, in a portable format.

WHERE IT RUNS

Works on every major web-based LLM chat interface. One lens, one set of features, same behavior everywhere. See the website for the current list of supported sites.

LOCAL-FIRST — your prompts never leave your machine

01. NO BACKEND, NO SERVER, NO TELEMETRY
aLLMost has no API of its own. The Free tier makes zero outbound network requests beyond the LLM platform you're already on. Pro license verification queries our database with your email address only, on popup open.

02. INDEXEDDB STORES EVERYTHING LOCALLY
Your prompt history, similarity vectors, and analytics live in your browser's local database. Cap is 100 prompts on Free, 1,000 on Pro. Clear it any time from the popup.

03. EVEN EVASION DETECTION RUNS LOCALLY
No external API call. Evasion % is derived from the same 300+ hedge/refusal/overclaim patterns that power the Free heatmap — classified per sentence, aggregated per response. Nothing about your prompts leaves the browser.

04. PERMISSION-LIGHT, MANIFEST V3
Two permissions only: storage (settings + license cache + IndexedDB) and activeTab (read the response stream from the LLM tab you're currently on). No CSP rewriting, no debugger access, no analytics SDK, no host requests beyond the LLM domains you already visit.

PRICING

▸ Free — Déjà Prompt + Confidence Heatmap + Evasion Detection. Forever. No timer.
▸ Pro — $7/month or $59/year. Adds Model Router, Quality Scores, Reuse Clusters, Personalized Routing, Prompts Export. Cancel from your Stripe billing portal any time.

LINKS

Website: https://allmost.app
Privacy Policy: https://allmost.app/privacy.html
Terms: https://allmost.app/terms.html
Support: Open an issue on the GitHub repo linked from allmost.app

LEGAL

aLLMost is an independent tool and is not affiliated with, endorsed by, or sponsored by any AI model provider. All product names and trademarks referenced are the property of their respective owners and are mentioned only for compatibility purposes.

aLLMost searches your past AI prompts by meaning and scores every response for hedging, evasion, and overclaiming — across major LLM chat platform, entirely in your browser.

Three layers, one extension.

BEFORE YOU SEND
Déjà Prompt watches your input field. The moment your draft semantically matches something you asked weeks ago — on any platform — it surfaces the original conversation. Stop re-asking your own questions.

AFTER THE RESPONSE
Every sentence is scored against 300+ hedging, speculation, and overclaim patterns. One evasion % per response. Read the linguistic shape of the answer, not just the words.

ACROSS EVERY CONVERSATION
Rolling 30-day quality scores per platform, derived from your own data. Reuse clusters surface the topics you keep cycling back to. Personalized routing calibrates to your taste over time.

FREE FOREVER, NO ACCOUNT

- Semantic prompt history — search your last 100 prompts by meaning, not keywords
- Confidence heatmap — 300+ patterns across four categories
- Evasion detection — per-response % score
- Works across all supported AI chat platforms
- Manifest V3, minimum permissions

PRO ($7/MO OR $59/YR)

- 10x prompt history (1,000 prompts)
- Model Router — intent classification + best-model recommendation across 8 task types
- Quality scores per platform with trend dashboard
- Reuse cluster analysis
- Personalized routing — your accepts, dismisses, and thumbs nudge the score
- Full prompt history export as JSON

WHAT THE HEATMAP IS AND ISN'T

The confidence heatmap scores hedging language, not factual correctness. A sentence saying "The capital of France might be Paris" tints red even when the fact is right. The signal correlates with hallucination but is not a truth check — it is a linguistic measurement. We are explicit so you can use it well.

YOUR DATA STAYS IN YOUR BROWSER

No backend. No telemetry. No outbound requests beyond the AI platform you are already on. Prompts, embeddings, and analytics live in IndexedDB. Even Pro features run entirely client-side. Clear or export your data any time.

HOW IT STAYS WORKING

aLLMost intercepts the network layer, not the rendered page. Network protocols are far more stable than UIs, so it survives the redesigns that break scraping extensions.

WHO IT'S FOR

- Power users running multi-LLM workflows
- Engineers, researchers, and analysts who need to know when a model hedges
- Anyone who has typed "I think I asked this before" into a fresh chat
- People who want their own data in a portable format

A product by Algomittens LLC.