Changelog & Roadmap Live

The emblem on the homepage is about a third larger, with a brighter glow, and the status light at its centre still reads /status. The sky behind it is darker and holds roughly twice as many stars, scaled to the screen so a phone is as full as a desk, with a few bright ones that twinkle. A ringed planet sits low on the left, lit from the emblem's side: far stars drift behind it and near ones pass in front. Shooting stars keep their pace and now trail a fading tail. Built for a phone first: the first button still shows without scrolling, and the sky holds still if your device asks for reduced motion.

The Daily Macro Pulse was written by Gemini and nothing else, so when Gemini refused, every retry failed the same way and the pulse could arrive hours past its 06:00 UTC slot. That happened on October 4, and on several mornings before it. It now tries Gemini first, then Claude, Grok and Groq, and the page says which model wrote that day's pulse. If the pulse is the only thing running late, the light on the homepage now says "Pulse delayed" instead of "Systems degraded". It still says LIVE only when everything on /status is operational.

The homepage now opens on two desks: Best Setups, where you pick a window and read a ranked board, and the trading skills an agent can install and run on paper. The light at its centre reads /status and says LIVE only when every system there is operational; otherwise it says degraded or down. Each skill's label comes from its live card. Built for a phone first: one button in the top bar, 44px tap targets, and the first button above the fold. The starfield holds still if your device asks for reduced motion.

After v0.9.5's fix, the grid reads its range from the first days of a test and DCA fixes its buy size before the first buy. Strategy Lab now lets you set both: "Range from first (days)", 20 by default, and "Per buy ($)", left blank to split capital across the scheduled buys. On phones its fields no longer make iOS zoom, and the Run button uses the lime primary style.

We measured each scoring input against what followed, on six days of live calls, scored the way the ranking scores it. The finding: the combined score does not yet put the board in a useful order on any window. On 1h, the biggest movers of the past hour did worse over the next one. Volume spike is the one input consistently on the right side, and that is not proven either; 1d has too few boards to judge. Nothing in the scoring changed in this release. The graded record on /setups stays the test, and its hit rate stays hidden until 30 graded calls.

Every window's board is recorded on a fixed schedule, from 1m every 15 minutes to 1y weekly, and each call is graded as a buy once its own window has run: close to close from stored candles, next to BTC over the same span. The record sits under the answer on /setups. A missing candle is never filled in; the call is marked unresolvable with the reason. A window's hit rate stays hidden until it has 30 graded calls, and every graded call, with the inputs that scored it, is public at /api/setups/track-record/calls.

The grid bot in Strategy Lab and the free 30-day backtest set its levels between the lowest and highest price of the whole test, so on day one it already knew the period's range. It now takes its range from the first days of the test and trades only after them. DCA fixes its buy size before the first buy, and cash not yet invested counts in the equity curve, so deposits no longer read as returns — DCA Sharpe ratios had been overstated. Grid and DCA results change; momentum and custom strategies do not.

The agent card and MCP tool list now say exactly what each paid skill reads. The "whale" feed is one public wallet's recent transactions, not market-wide transfers over 500 ETH, and the strategy tool lists the four strategies it runs instead of three it never had. Agent operators start at /agents.

A paid MCP tool ran for any transaction hash it was handed, so a made-up one ran a $1 tool for free. Paid tools now run only for Premium or a payment confirmed on Base, to us, for the price, and not spent before.

Three trading skills with fixed, hashed rules, each with a card: a backtest before its listing date and a forward record after, wins counted after fees. Agents can pay with standard signed x402 payments, settled through PayAI and read back from Base before anything is served. Pro and Premium builders can upload paper-only skills, listed as unverified. Assistant Hub is listed on the official MCP Registry. Details on /developers.

From September 8 the shared nav was missing from every public page in production. It is back, and phones get a bottom tab bar: Setups, Pulse, Research, Console and More. In the app, crypto tabs, inputs and toggles meet the 44px tap size, inputs no longer make iOS zoom, and paid gates name the plan they need.

A board built moments ago is served at once while a fresh one builds — at most 90 seconds old on 1m, 15 minutes on 1w and 1y — and the card prints its real as-of time. The candle API had been adding up 24-hour volume snapshots to build hourly and daily candles, so one daily BTC candle claimed more than half of all bitcoin. A merged candle now carries its closing figure, the API says what its volume field means, and the app's chart plots ticks.

Every setup on /setups has Arm watcher, which opens /watch prefilled with that setup: a price band scaled to its window, both levels shown before you arm, ending in Telegram pairing. A free account now holds one watcher — the loop no longer stops at a paywall — and more is still an upgrade.

All eight setups on the 1-week board sat inside two points. The weekly return capped at +15%, so every strong week scored the same, and trend quality could only take a couple of values. The return now spans the range markets actually move in, and trend quality measures how cleanly the week got there — a clean climb scores high, chop scores low, a fall scores zero.

The window chips sit at the bottom edge, in thumb reach. Setups is in the app's bottom bar, and More opens on phones. Cards say when price is outside its Bollinger bands, and the risk advice matches the level printed beside it.

The homepage now opens on pick a window, get ranked setups. "Set a Telegram alert" goes to the watcher instead of the Strategy Lab backtester, and NEW badges are gone from everything off that path.

Invite-only rooms no longer show up to outsiders, archived rooms stop working, and wallet-gated joins sign a one-time challenge instead of a reusable string. Changing your password signs out your other sessions, and sign-in signatures go through an audited library. A card checkout can no longer open a second plan, and USDC payment callbacks take their price from the server, not the request. Details on /security.

The Daily Pulse no longer claims a Grok agent or on-chain data it never used. Public pages now match the product, including that card plans renew monthly with a one-click cancel. Removed the voice button — our own microphone policy blocked it — and a static "agents online" ticker that reported nothing live.

/setups: choose 1m, 1h, 1d, 1w or 1y and get up to eight deterministic, ranked setups — each with its score inputs, why now, what would invalidate it, a risk posture and an as-of time. /api/setups/health reports which windows can rank and why not. Stablecoins are never ranked.

Google sign-in only accepts tokens issued to this app and no longer merges into password accounts. Persona memories are private to their owner. /metrics no longer opens to any Premium account, and webhooks cannot target internal addresses. Session tokens are kept out of page JavaScript, and user-supplied text is escaped across the forum, rooms, signals and audit log.

The hash chain behind /audit-log is now written in one serialised database transaction, so concurrent writes and deploys no longer break it. The log no longer shows usernames, user ids or wallet addresses.

An on-chain payment is accepted only once a Base RPC verifies it. USDC plan payments are paused until your own wallet signs them; card payments work.

Removed entirely, including the analysis endpoint that could charge for a request it could not run. Whale analysis now tells the model exactly what on-chain data it has.

The agent endpoint speaks A2A 1.0 alongside 0.2/0.3, chosen per request, with v1 agent-card fields and an /agents catalog. A2A tasks are visible only to their owner.

Visual no-code builder inside Arena/Strategy Lab — choose strategy type, write a system prompt, set model/interval/risk. Three tabs: Build, Templates, My Agents.

Four ready-to-deploy templates: Whale Alert Tracker, RSI Sniper, Sentiment Radar, and Macro Hedge — one-click fork and customize.

Bell icon in the header with unread count badge. All system alerts, agent triggers, and price alerts in one panel with read/unread state and digest settings.

Configure webhook URLs and email addresses. Get agent triggers and price alerts wherever you are. Pro/Premium feature via dedicated Settings panel.

Dismissible banner on first login after each release — highlights the most important new features with a direct link to this changelog.

Browser-native speech recognition for hands-free chat. Microphone icon in the input bar; auto-submits on pause. Works with all AI providers.

Smart install banner after 3+ visits, updated manifest with app shortcuts, offline fallback page, and splash screen for all screen sizes.

Updated 7-step tour now covers Agent Builder and Notifications. New "Take a Tour" button in the What's New banner.

Developer-facing API key management UI in Settings. Manage access tokens for programmatic agent deployment and signal retrieval.

Community Feed shows curated signal highlights and agent leaderboard updates. Strategy Lab backtesting reliability improvements and clearer result cards.

This page. Full timeline with animated dots, tag filters, version badges. Landing page top nav (Features dropdown, Arena, Pulse, Changelog). Sidebar Changelog link in Account group.

Live Tips cheatsheet, critical badge pulsing, pinned group elevation, drag-to-reorder groups, persistent search, Ctrl+Shift+E global toggle, sidebar analytics dashboard.

Sortable live table, threat scoring (0–100), wallet tracking, DeFi flows, CSV/PDF export, interactive heatmap, per-wallet sparklines with 24h/7d toggle.

Three.js and React removed from bundle. −1.4 MB total, −71 console errors, 40% faster cold starts.

Fuzzy-search launcher for pages, theme toggle, focus mode, settings. Keyboard-first navigation across the full app.

3-agent pipeline: Sentiment (Grok) + Technical (Groq) run in parallel, Synthesis (Gemini) combines results.

Server-side backtesting (DCA, Grid, Momentum, Custom), virtual wallet, simulated fills with volume slippage, stop-loss/take-profit, PnL dashboard.

Auto-generated at 06:00 UTC — top 3 macro threats + top 3 crypto opportunities, 7-day archive, public /pulse page.

Collaborative analysis rooms where multiple AI agents and users research together in real-time.

NIST SP 800-53 audit logging, Agent Capability Attestations (default-deny), AES-256-GCM key encryption, distributed request tracing.

Live prices, 24h change, volume, market cap for top 100 coins. Chart timeframes (1D/1M/3M/1Y/5Y/All), 5-min server cache.

Live Fear & Greed panel (alternative.me) and CryptoCompare news feed — no API key required.

MetaMask, Coinbase Wallet, OKX, Phantom, Rabby — 4-layer detection fallback, fixed Coinbase recognition.

6 learning tracks: Web Dev, TypeScript, APIs, Crypto, AI Engineering, Ask Anything.