adderall mascot

ADDERALL

A local-first AI agent that lives in your browser. It thinks step-by-step, uses 44 tools, scans Solana tokens, tracks wallets, builds websites, learns from every conversation, and decomposes complex tasks — all with persistent memory that gets smarter over time.

Contract Address TBA Click to copy

FEATURES

🧠

Agentic Reasoning

THINK → ACT → OBSERVE → DECIDE loop with chain-of-thought. Up to 24 self-extending steps per request. Goal decomposition breaks multi-phase tasks into subtask chains.

🔍

Crypto Intelligence

Live token prices, DexScreener charts, top holders, risk scores, fresh wallet flags, multi-token comparison, cross-referencing, wallet PnL tracking with visual cards.

🌐

Build & Deploy Sites

Describe a website, get full HTML/CSS/JS. Edit with follow-ups. Deploy to Netlify. AI critic auto-scores designs and fixes anything below 8/10.

📚

7-Layer Memory

Condensed history, recent chat, user profile, session summaries, manual memories, learned skills, and project plans — all persisted with relevance-filtered injection.

Self-Evolving Skills

Learns behavioral patterns from your conversations. Skills scored 0–100, evaluated by LLM + implicit satisfaction signals, auto-pruned when ineffective.

🤖

Autonomous Tasks

Background work, recurring monitors, price alerts, reminders, and daily digest. Each runs through the full agent loop and delivers results via notifications.

👁

Page Awareness

Sees your current page, tweets, DEX pairs, wallet addresses. Auto-detects DexScreener, Birdeye, pump.fun, Jupiter, and Solscan pages.

💰

Wallet Tracking

Track up to 20 wallets with labels. Auto-polls for buys, sells, and large SOL moves. PnL cards, daily digest, and configurable USD thresholds.

💊

Companion Mode

Floating pill mascot follows you across every page. Drag anywhere, hover for quick-chat, see background results in the bubble. Click to open full panel.

44 BUILT-IN TOOLS

Web SearchReal-time via DuckDuckGo
Fetch URLRead any webpage
ScreenshotCapture & analyze pages
Reverse ImageIdentify people & memes
Token ScanHolders & concentration
Multi-ScanCompare up to 5 tokens
Token PriceLive prices via Jupiter
DexScreenerCharts, liquidity & volume
Wallet BalanceSOL + token holdings
Wallet HistoryRecent transactions
PnL CheckFull portfolio report
PnL CardVisual shareable snapshots
Watch WalletPersistent watchlist
Wallet ActivityTracked buys & sells
Daily DigestScheduled briefings
Detect MintsExtract Solana addresses
List ScansAll scanned tokens
Build SiteGenerate full websites
Edit SiteModify with prompts
Deploy SitePush to Netlify
List SitesManage Netlify deploys
Delete SiteRemove old deploys
RememberSave to persistent memory
RecallRetrieve saved info
Search MemorySemantic embeddings search
Set TimerDelayed background tasks
Schedule TaskNon-blocking background work
MonitorRecurring watch loops
List / StopManage active monitors
Set AlertPrice & wallet alerts
Remind MeSmart reminders
ContinueSelf-extend reasoning
Learn SkillSave behavioral skills
List / ForgetManage learned skills
Set GoalMulti-session projects
Prepare LaunchPump.fun auto-fill

HOW IT WORKS

  1. Install — download from GitHub, load as unpacked extension in Chrome
  2. Configure — add your LLM API key (OpenAI, Anthropic, or compatible) and optional Solana RPC in Settings
  3. Use — click the adderall icon on any page. Ask questions, scan tokens, build sites, track wallets
  4. Right-click — send text, links, or images to adderall from the context menu
  5. It learns — the agent evolves skills from your interactions, scores them, and prunes what doesn't work
  6. It works alone — schedule monitors, alerts, and background tasks that run through the full agent loop
ENGINE

POWERED BY THE LOCKEDIN ENGINE

Locked In is the agentic reasoning engine behind adderall. A deterministic THINK → ACT → OBSERVE → DECIDE loop with 7-layer persistent memory, semantic search, self-evolving skills, goal decomposition, implicit satisfaction signals, and anti-context-rot defenses. Not a chatbot wrapper.

WHAT MAKES IT SMART

Goal Decomposition

Multi-phase requests are auto-planned into subtask chains with dependency tracking, each running through the full reasoning loop.

Satisfaction Signals

Detects corrections, frustration, and praise in your messages to nudge skill scores — the bot gets smarter from how you react.

Self-Extending Reasoning

Can expand its own step budget up to 24 steps for complex analysis instead of cutting short.

Context Rot Prevention

Relevance-filtered memory injection, caps, pruning, condensation, and an 80K char hard budget keep the context window clean.

Proactive Briefings

Opens with price updates, wallet balances, and active alert summaries when you have monitors running.

Self-Critique on Builds

AI critic scores site designs 1–10 and auto-fixes issues below 8 before showing you.