🔎
Agent: Default Kaizen
Kaizen
Attach images or video, or just ask — generated media renders right here in the conversation.
🎙️
Tap the mic to start a live conversation
Hands-free: say "Hey Kaizen" once, then keep talking — commands, dictating a message, whatever you need — without repeating it every time. Kaizen only acts on things said to it, so a conversation with someone else in the room won't trigger it.
Voice engine: checking…
Transcript will appear here...
Turns this session
0
Time listening
0s
UI commands
0
"Hey Kaizen, open [tab name]" — jump to any section
"Hey Kaizen, open my files" — jumps to Workspace/Files
"Hey Kaizen, open Minecraft" / "walk forward" — routed to the engine's screen_control capability on your device (needs Screen Control enabled)
"stop talking" — cancel speech playback
"screenshot" — routes to screen control
"enable/disable autonomous" — toggle mode
Live Shell Sessions checking...
A real shell running on the server, not a chat round-trip -- run
vim, top, ssh, long-lived processes, whatever you'd run in any terminal. Open several tabs for several independent sessions at once; each keeps running (and remembers its scrollback) even if you switch away or close the tab in your browser.No live session open. Click "+ New session" to start one.
Raw Engine Console ↑ / ↓ history · Ctrl+R search
Sends whatever you type straight to the engine's
/chat endpoint and prints the raw response, REPL-style. Useful for exact commands like capability_health, list capabilities, or direct screen_control syntax. This is NOT a real shell -- for that, use Live Shell Sessions above.Kaizen terminal ready. Type a command and press Enter.\n
kaizen>
Live Background Jobs 0 running
Long-running tasks that keep working over time (capability creation, self-edits) - started from
chat/terminal and polled here live, even after the request that started them already returned.
No background jobs running.
Task Queue 0 queued
Tasks dispatched this session, newest first.
0 done
0 running
0 failed
No tasks dispatched yet this session.
Registered Capabilities –
Live from
list capabilities via the engine.Click refresh to load capabilities from Kaizen.
Raw Capability Health
Click refresh to pull capability_health from the engine.
What this does
A workflow is an ordered list of chat commands you save once and re-run as a single button press. Each step is sent to Kaizen exactly like typing it in Chat, one after another, in order — step 2 doesn't start until step 1 returns. There's no branching or conditionals; it's a straight-line pipeline, so it's best for routines you'd otherwise type manually every time (a daily check-in, a report pulled from several sources, a cleanup pass).
Example: a 3-step "Morning Report" workflow —
system statussummarize anything important from overnight notificationslist capabilities
Running it fires those three, in that order, and each step's row shows idle → running → done/failed live as it goes.
Template Library
One click loads a ready-made multi-step pipeline into the builder below — tweak and save.
Build a Workflow
No workflows saved yet. Build one above.
Run History 0
Every workflow run this session, most recent first.
No workflow runs yet.
What this does
Automations repeat one command on a fixed interval (e.g. every 30 minutes) for as long as this tab/session stays open, logging each run and notifying you of the result — good for recurring checks like "capability_health" or "summarize new emails".
Scheduler below fires a command once at a specific date/time — good for a one-off reminder or a task that only needs to happen at a particular moment.
Device control: if you enable the global "Device Assist" toggle (top bar) and check "Allow device control" on an individual automation, that automation's runs are allowed to use this browser's device bridge (screen/clipboard/local actions) the same way a manual chat message can. It's off by default per-automation even when the global toggle is on, so recurring unattended jobs don't get device access unless you explicitly opt each one in.
Example: name "Health Check", command capability_health, every 60 minutes, device control off.
Template Library
One click pre-fills the New Automation form below — tweak the interval and create.
New Automation
min
Daily Command Load
Estimates how many command runs your active automations will fire per day, and flags any pair scheduled to fire within the same minute (likely to queue behind each other).
Active Automations 0
No automations yet. Create one above — e.g. run "system status" every 30 minutes.
Automation Run Log
No runs yet.
Scheduler (one-off tasks)
Schedule a Task
Upcoming 0
Nothing scheduled yet.
30-day trend
–
Projected next 30d
$0.00
Slope ($/day)
–
⚡ Revenue Focus Mode
Runs Kaizen on this device's own power only (no hive) and pushes every revenue
automation to its fastest safe cadence. Nothing bypasses approval or spends money
on its own -- it just makes the existing free/zero-risk automations run harder.
Focus Mode
Pending review
0
Approved, unexecuted (stale ≥24h)
0
Needs you (no auto-execute path)
0
Real income (verified)
$0
Claimed, not yet verified
$0
Needs verification
0
Referrals credited
$0
Best avenue ($/hr)
–
Live trading P&L
$0
"Real income" only counts money Kaizen has actually confirmed landed (Coinbase balance growth, or
your own manual verification below) — not just claimed. "Claimed, not yet verified" is every dollar
logged from a marked-done lead that nobody has confirmed actually arrived yet; treat it as a
pipeline, not money in hand.
Completion Board
Every job's real progress, not just "generated": to-do → in-progress → needs you →
blocked → done. The background completion engine actively pushes stuck jobs
forward or tells you exactly what it needs — nothing sits silently stale here.
To Do
In Progress
Needs You ●
Blocked
Done ●
Second Chances rejected leads that were worth real money -- revive for another look
Growth Signals who to re-engage and what to pitch, from proven buyer history
Churn risk repeat buyers overdue
Cross-sell next avenue to pitch
Reactivation one-time buyers, worth a nudge
Best outreach template by realized close rate
Revenue by channel $ + close rate per source
Avenue scan freshness avenues the scan cycle hasn't visited lately
Revenue Goals
Financial Snapshot
Real vs pending honest split of every $ recorded
Tax set-aside rough estimate, not filing advice
Top wins biggest executed payouts
Zero-Risk Opportunities research studies, bug bounties, beta programs -- no approval pipeline, just apply directly
Focus avenues –
Kaizen scans these avenues most often (free, most autonomous, most reliable); every other avenue still gets swept periodically so a strong lead elsewhere isn't missed.
Income Goal Tracker
Manually log verified income against a goal to see progress. Saved locally in this browser.
$0 of $1000 this month (0%)
Lifetime logged: $0 · milestones unlock as it grows
Lead ROI Calculator
Sanity-check a lead before approving it — figures out the real $/hr after fees, and how many runs it takes to break even on any upfront cost.
Quarterly Estimated Tax (rough estimate — not tax advice)
Simplified U.S. self-employment estimate: 15.3% self-employment tax plus a flat federal rate you set, applied to net profit. Doesn't account for deductions, state tax, or the SE tax deduction — use a real accountant for filing.
Task type × status breakdown
Every lead grouped by task type (kind) down the side and completion status across the top. Click any count to jump to that exact slice below -- a clean cross-section instead of one long mixed list.
| Task type |
|---|
| Loading... |
Leads & approval queue 0
⏳ Needs your decision 0
Nothing here is being worked on -- Approve/Reject before it does anything.
No revenue leads yet -- Kaizen scans in the background and they'll show up here.
🔄 Already approved / in progress 0
You already said yes to these -- shown separately so they don't get lost in the decision queue above.
Nothing in progress right now.
🔍 Other lead queues (bounties / affiliate / freelance) 0
These are found and stored separately from the queue above -- scanned on their own schedule, so they didn't previously show up here even though Kaizen was finding them.
Loading...
Paste a job brief you found yourself (Kaizen doesn't browse job boards for this one) and it'll draft a proposal and a short approach outline for you to review. Approve, reject, or ask for a revision -- nothing is ever sent to a client automatically.
No pending drafts.
Data quality 0 flagged
Catches three specific problems automatically: leads whose avenue got mislabeled (typo'd or a classifier hallucinating a key), spam/marketing emails that got ingested by an email-scanning avenue and turned into a fake "lead", and leads that are entirely fabricated (unreplaced placeholder text, a target URL pointing at a dead/example host, or payout economics that don't describe anything real). Mislabeled items get their avenue field corrected in place; spam and duplicates get rejected (never a hard delete -- they still show up under Rejected if you want to double-check a call). Fabricated leads are fully removed from every list (not just rejected) since there's nothing real to review -- restorable from the Hard-deleted section below if a call was wrong. Only ever touches pending/approved items, never anything already executed.
Muted avenues
— new leads from these get auto-rejected instead of piling up
🗑 Hard-deleted (detected as fake leads)
— fully removed from every list, not just rejected; restorable
Loading...
Recent cleanup runs
No runs logged yet.
Duplicate detection 0 pairs
Scans pending items for likely duplicates by avenue + text similarity. Dismissing a duplicate rejects it and links it to the item you're keeping -- it never silently deletes anything.
Click Scan to check for duplicates.
Profitability & Forecast –
Net = executed-item revenue minus logged costs, per avenue. Forecast is a simple trend line over recent daily earnings -- directional only, not a promise.
Total gross
$0
Total costs
$0
Total net
$0
Bleeding avenues
0
Projected next 30d
$0
| Avenue | Gross | Cost | Net | ROI |
|---|---|---|---|---|
| Loading... | ||||
⚠ Bleeding avenues
-- net_usd at or below the threshold; informational only, nothing here auto-mutes
Not checked yet.
Log a cost
Taxes bookkeeping only, not tax advice
No earnings recorded yet.
Self-sustaining reserve balance
$0
Affiliate revenue (all-time)
$0
Sponsorships pending review
0
Storefront revenue (all-time)
$0
B2B API keys active
0
B2B billed (all-time)
$0
Verified real income
$0
Pending review
$0
Transfer requested, not yet verified
$0
Run-rate (trailing 30d, straight-line)
$0
Recurring share of trailing income
Referrals credited (all-time)
$0
Trailing-30d income by category
No data yet.
Referral leaderboard
No referrals yet.
Affiliate links — top performers
| Program | Product | Clicks | Conv. | Rev/click | Revenue |
|---|---|---|---|---|---|
| No data yet. | |||||
Digital storefront — top products
| Title | Platform | Sales | Revenue | Status |
|---|---|---|---|---|
| No data yet. | ||||
Pricing suggestions A/B-test signal only — nothing here changes an external listing automatically
No signal yet — needs at least 5 sales on a listed product.
B2B API tier signals
Near quota — upsell candidates
None right now.
Underused — churn-risk / reclaimable
None right now.
Margin health — flag keys priced too close to actual per-call cost
Enter an estimated cost/call and check.
Pending sponsorship offers
None pending.
Accounts running
0
Best performer
–
Total trades
0
Profitable accounts
–
Global champion
–
Fear & Greed
–
Avg return (all)
–
Strategy / Asset accounts 0
| # | Asset | Strategy | Equity | Return | Max DD | Sharpe | Trades | Win rate | Position | Started | Status | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Not running yet. | ||||||||||||
Strategy Pool the current best 12 of 20 strategies, ranked by real paper-trading performance across every asset (risk-adjusted) TradingView: checking…
Loading…
Regime Performance which strategies actually do best by weekday -- feeds a live ranking bonus, shown here for the first time
Loading…
Correlation Matrix pairwise return correlation across every active-pool asset -- same math the promotion/drift guards already use, shown directly for the first time
Loading…
Sector Concentration how much of Kaizen's currently-open exposure sits in each asset sector -- now a live entry-sizing gate, not just a readout
Loading…
Signal Confluence every independent signal Kaizen already checks at trade time (TradingView, daily trend, VIX, fundamentals, volume) for each active stock, side by side for the first time
Loading…
Kaizen's Day-Trading Research Journal what it's actually been learning, not just what it's trading
Loading…
Max Drawdown Calculator
Paste an equity curve (comma or newline separated values) — e.g. copied from a backtest — and get peak-to-trough drawdown stats. Client-side only.
Compounding Growth Projector
Pure math projection from a win rate and average win/loss size — not a prediction, just "if this edge holds, here's the curve." Client-side only.
| Period | Expected equity | Growth |
|---|---|---|
| Enter figures and hit Project. | ||
Position Sizing & Risk Calculator
Client-side only — figures out share/coin size from your account risk %, and the resulting risk:reward ratio. Doesn't place or size any real order.
Trade Journal 0
A quick manual log of wins/losses/breakevens outside the automated strategies — for your own discipline tracking. Saved locally in this browser.
Win rate
–
Current streak
–
Total logged
0
Avg R-multiple
–
Live trading status checking…
LIVE_TRADING_ENABLED is an environment variable set by hand on the server -- nothing here can flip it. Even when it's on, an order only goes through for a promoted (asset, strategy) pair, at or under the capital limit you set below.Global switch
–
Coinbase connection
–
Max per trade
–
Max daily loss
–
Live capital committed
–
Promotion candidates 0
Standing paper-trading champions. Promoting one lets it place real Coinbase orders, capped at whatever $ amount you approve here -- it never sizes itself. Demote at any time to immediately revoke real-money authorization (the strategy keeps running on paper regardless).| Asset | Champion strategy | Return | Max DD | Trades | Champion for | Status | Action |
|---|---|---|---|---|---|---|---|
| No champions yet -- let the paper tournament run for a while. | |||||||
Live P&L by strategy $0.00
Real, closed P&L from actual Coinbase fills -- separate from the simulated paper-tournament returns above. Empty until a promoted strategy closes its first live trade.| Asset | Strategy | Realized P&L | Closed trades | Win rate | Open position |
|---|---|---|---|---|---|
| No live trades yet. | |||||
Live Trading Risk Report –% of daily cap
Today-only realized P&L per live pair against the daily loss kill threshold -- catches an account-wide cap breach that no single pair shows alone.Loading…
| Asset | Strategy | Today's realized P&L | Lifetime P&L | Largest single loss | Win rate | % of daily cap |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Drawdown vs. auto-demote threshold
(peak-to-current equity, auto-demoted if breached)
| Asset | Strategy | Peak equity | Current equity | Drawdown | Distance to auto-demote |
|---|---|---|---|---|---|
| Loading… | |||||
Quick Templates
One click pre-fills the form below — tweak and create.
New Agent
No custom agents yet — the default Kaizen persona is active.
Friends
0
Incoming requests
0
Outgoing requests
0
Add a friend
Send a friend request by username. They'll see it under Incoming and can accept or decline.
Incoming requests 0
No incoming requests.
Outgoing requests 0
No outgoing requests.
Friends 0
🔎
No friends yet — send a request above, or invite one from a game lobby.
Blocked users
Blocked users can't send you friend requests. Blocking also removes any existing friendship or pending request.
No blocked users.
Deep Research Runtime
Multi-source analysis, autonomous reasoning, recursive decomposition, and live synthesis pipelines. Ask a research question from the Chat tab and Kaizen routes it through
web_research.Under the hood, web_research queries the DuckDuckGo instant-answer API plus an HTML-search fallback (so a quiet/empty instant answer doesn't mean a quiet result), pulls in up to 8 sources, and has the model synthesize them into a plain-prose briefing rather than handing you raw search snippets. This is specifically what lets Kaizen answer things past its own training cutoff — current events, "is X still true", "does Y support Z now" — instead of guessing from stale knowledge. It's also what capability-creation and self-editing use internally to research an unfamiliar library or API before writing code against it.
Example: asking what changed in the most recent stable release of the Linux kernel will pull current sources and come back with a synthesized summary + the sources it drew from, instead of an answer frozen at whatever the model last learned.
Quick Research 0 queries this session
Fires a research request directly.
Research Output
Results will appear here.
Saved Research History 0
🔎
Nothing saved yet.
Upload & Generated Artifacts
Drag & drop images or videos here, or use the upload button. Files are sent to the engine's
/upload endpoint — see setup notes below if that route hasn't been added to your backend yet.
Drop images / videos here, or click to browse
0 files · 0 KB this session
🖼️ Images: 0 KB
🎬 Video: 0 KB
Backend setup required action needed
This UI posts uploads to
POST /upload (multipart/form-data, field name file) and expects
{"url": "/workspace-files/<name>"} back. It also expects generated artifacts (images/video from
generate_image etc.) to be reachable at a static route. See the chat reply for the exact FastAPI
snippet to add — this can't be faked from the frontend alone.
Open a project
Upload a
.zip of a codebase to browse, edit, and run it right here -- or jump straight to a
project you already uploaded by pasting its ID. Uploads go to POST /api/upload-project and are
saved under workspace/projects/<id>, the same directory project_improve works from.
or
or
Live Edit -- watch Kaizen work on a file
Upload any single file, tell Kaizen what to do with it in the panel that opens on the right, and watch it
work in real time -- step through what it did, pause, and send new instructions mid-run.
Honest caveat: this couldn't be smoke-tested in a real browser here, only checked statically
(syntax, matching IDs, backend logic run against real files) -- worth trying it yourself before relying on it.
Long-Term Memory
Backed by
knowledge_base / memory.json on the engine — conversation history, learned capabilities, and mutation history all persist across restarts.Knowledge Graph
Relationship mapping between goals, workflows, agents, research tasks, and learned information.
Memory Search
Engine Status
Click refresh to pull the full status report.
Knowledge Base (reference docs)
🔎
Add a Note / Document
No documents yet.
Quick Notes (server-persisted)
New Note
Short, taggable notes stored on the engine — survive restarts, separate from conversation memory and the reference docs above.
Notes 0
No notes yet.
Screen Control requires KAIZEN_ENABLE_SCREEN_CONTROL=1
Sends natural-language screen-control commands straight to the engine's
screen_control capability. Say the same commands from the Voice tab and they'll execute exactly the same way.Screen control output will appear here.
🎙️ Hands-Free Mode off
A dedicated mic that only drives the screen -- it never goes to chat. Say a whole chain in one breath, e.g.
"go on google, search youtube, hit enter" -- it's split into ordered steps and run one at a time,
at a pace that gives each step (page loads especially) time to actually land before the next one fires.
Most-Used Commands tracked this session
Your top commands so far — click to run again instantly.
Run a few commands and your most-used ones will show up here.
Command History 0
No screen control commands run yet.
Live Screen Watch browser capture
Uses your browser's own screen-share permission (not the OS-level engine) to visually watch what's happening while a mission runs — handy for keeping an eye on long
screen_control automations. Captures go to the Cross-Tab Monitor tab as well.Broadcast to All Tabs
Sends a message that pops up as a notification in every other open Kaizen tab — handy for coordinating work across windows.
Rename This Tab
Cross-Tab Activity 0 total events
Live Task Feed 0 events
Task start / finish / failure events broadcast from every tab, including this one.
No cross-tab activity yet.
Screen Watch Captures 0
Frames captured from the Screen Control tab's live screen watch (this browser tab only — capture streams aren't shareable across tabs for privacy/security reasons in the browser sandbox).
Games
The same evolving story and world for everyone — plus games you can customize yourself. Visit the Friends tab to invite people to play.
🎉 Party — carries between games; creating a lobby auto-invites your party
You're not in a party. Create one, then invite friends -- whenever you create a lobby in any game, everyone in your party gets auto-invited.
🙂 Your Avatar
Your 3D character, used automatically in every game. Drag the preview to rotate it, and pick your skin tone and outfit color below.
Avatar style
Skin tone
Hair style
Hair color
Outfit type
Outfit color
Eye shape
Eye color
Glasses
Mustache
Hat
Face
Accessories
🖌️ Custom sprite design (2D)
draw or upload your own pixel art for the 2D avatar — as detailed as you want
⠿
✨ AI Generate
Detail:
Zoom:
100%
Color
16×16 grid · click and drag to paint
This design is your whole 2D character — there's no separate body underneath it.
Pastes this onto the 3D Stylized outfit as its own texture. It's a one-time copy, like a Minecraft skin — the 3D look stays independent after that until you press this again.
🧑🎤 Hyper-Realistic Avatar Beta
🐾 Companion Pet
Species
Color
Pattern
Accessory
Aura
Trait
🎭 Party Games — Mafia, Werewolf, Spyfall, and more, with friends
Loading party games…
🎲 Board & Card Games
Loading board games…
⚡ Quick Play
Reflex Tap no runs yet
No loading, no backend — just click Start, then tap the target the instant it turns green. Best time is kept on this device.
Memory Match no runs yet
Flip two cards at a time and find every pair in as few moves as possible. Best score is kept on this device.
Whack-a-Mole no runs yet
30 seconds on the clock — tap moles the instant they pop up. Best score is kept on this device.
Quick Math Sprint no runs yet
60 seconds — solve as many quick arithmetic problems as you can. Best score is kept on this device.
Number Merge no runs yet
Swipe or use arrow keys to slide tiles — matching numbers merge into their sum. Reach the highest tile you can. Best score kept on this device.
Tap the board then use arrow keys, or use the buttons below on touch.
Simon Says no runs yet
Watch the color sequence, then repeat it back. Each round adds one more step. Best round kept on this device.
Tic-Tac-Toe no runs yet
Play against the computer. Fully offline, no loading. Best win-streak kept on this device.
Snake no runs yet
Arrow keys / WASD to steer, or swipe on touch. Eat the food, don't hit yourself or the wall. Best score kept on this device.
Offline Games — Stats Hub
Your best scores across every offline mini-game on this device, in one place.
🎮 Games
🔎
Click any game to load it full-screen — real controls, real matches, expandable on mobile and desktop.
Loading…
🛠️ Sandbox — describe a game, Kaizen builds it
Game
Kaizen runs headless, so it can't ask "apply this change? [y/N]" interactively — every self-improvement or self-edit proposal that passes all of its own safety gates (validation, protected-symbol guard, critique, sandbox smoke test) still lands here for a human to review before it's ever written to disk. Approving re-runs those same gates one more time; it only skips the interactive prompt, not the safety checks. Applying a change to Kaizen's own core files requires a restart to take effect, which happens automatically on success.
Pending review
–
Approved (all-time)
–
Rejected (all-time)
–
Gate pass rate
–
Safety gate pipeline every proposal passes through in order
A proposal only reaches this queue after clearing all four gates below — approving doesn't skip them, it re-runs them once more before writing to disk.
① Validation
② Protected-symbol guard
③ Critique pass
④ Sandbox smoke test
→ awaiting human approval
Loading…
Every approve / discard / rollback decision, newest first. Approved changes that were actually written to disk can be rolled back here to their exact pre-edit contents.
Loading…
What people keep asking Kaizen to do that no existing capability matched, tracked automatically every time a request falls through to capability-learning. The background learning loop already prioritizes these over curated topics — "Teach me this" just jumps the queue and runs it right now, through the exact same gated self-improvement pipeline as typing the request in chat.
Loading…
A repaired capability staged as a gradual rollout instead of an immediate cutover: it runs silently alongside the live version first (shadow), then earns a small slice of real traffic (canary), and either promotes itself to fully live or auto-reverts — whichever the observed fail rate actually earns. Nothing here needs babysitting; refresh just shows where each one currently stands.
Loading…
What happened the last time each staged repair finished — promoted to live, or reverted with the fail-rate comparison that triggered it. Same record start-to-finish as the Self-Edits history above, just scoped to canary rollouts specifically.
Loading…
Describe a change to this page in plain English — Kaizen edits index.html directly through the same gated mutation pipeline as every other self-edit above (generation, danger-pattern scan, critique, then this same human-approval queue for anything risk-flagged). "Preview first" runs the whole pipeline and shows you the diff without writing anything to disk; uncheck it to apply immediately if the change clears the gates cleanly. Either way, reload the page afterward to see the result.
Total accounts
–
Custom permission sets
–
Suspended
–
Trusted devices
–
0 selected
Loading…
Permission presets 0
Named bundles of capabilities + tabs you can apply to any account in one click, instead of hand-picking checkboxes every time. Built-ins cover common roles; add your own below and they'll show up here and in the bulk-action bar.
User tags 0
Free-form labels for organizing accounts (e.g. "contractor", "family", "on leave") -- attach them from any user's row. Purely local to this browser, doesn't touch permissions by itself, but makes filtering and bulk selection much faster.
Permission audit log 0
Every grant, revoke, suspend, delete, and preset application made from this browser, newest first. Kept locally so you have a trail even if the server doesn't log admin actions itself.
Invite keys 0 active
One-time signup keys that let someone create their own account without you creating it for them by hand -- they still start with zero permissions (Chat only) until you grant tab/capability access above. A key is consumed on first use, or expires on its own if you set a time limit.
Loading…
Access Lock checking…
Restricts who is allowed to reach this hive node's endpoints at all,
checked before any hive-secret verification. See
hive.py lock-tailnet /
rig_agent.py lock-tailnet — this panel is a front-end for the same config.Lock status will show here.
Connectors 0 connected
Third-party accounts Kaizen can read from or act through once connected (e.g. email, calendar, task trackers, social). Toggling one on here doesn't grant access by itself — it flags which services Kaizen should route matching requests to; you still complete that service's own login/authorization the first time it's actually used.
Webhook Tester
Fires a test POST to any URL you provide (subject to that server's CORS policy). Use this to confirm an endpoint is reachable and responds the way you expect — for example, checking that a Zapier/Make webhook, a Slack incoming webhook, or your own server route accepts Kaizen's payload — before wiring it into an automation or workflow step for real.
Result will appear here.
Recent test sends:
API Explorer
A raw client for Kaizen's own backend API (this server, not a third party) — useful for testing an endpoint directly, debugging a response, or building your own script/integration against Kaizen. Pick a method and path (e.g. GET
/api/health, POST /chat with a JSON body) and Send fires the request exactly as written; the raw response, including errors, shows below.Request
Response
Response will appear here.
Registered Capabilities
Click Refresh to load capability availability from the engine.
Run a One-Off Test Call
Invoke any registered capability directly, bypassing chat routing -- useful for confirming a tool actually works before relying on it.
Output will appear here.
Total Executions
0
Success Rate
100%
Avg Response Length
0
Session Duration
0s
Executions per Minute (last 10 min)
Breakdown
Messages sent0
Messages received0
Attachments uploaded0
Voice interactions0
Screen control commands0
Automations executed0
Workflow runs0
Terminal commands0
Daily Streak 0 days
Consecutive days you've opened Kaizen. Keep the streak alive!
Achievements 0/0
Milestones unlocked as you use Kaizen.
Weekly Activity Heatmap
When you're actually active, by day and hour, over the last 4 weeks — entirely tracked on this device.
Less
More
Not enough data yet — keep using Kaizen and this fills in.
Personal Records
Best-ever numbers, computed from your local streak and activity history.
Longest streak ever
0 days
Total active days
0
Busiest hour
–
Busiest day of week
–
Export
Download your usage stats as a snapshot.
Focus Timer idle
A simple Pomodoro-style timer to structure work sessions. Runs entirely client-side.
25:00
Sessions completed today: 0
All systems nominal
Refresh to run a live diagnostic pass.
Total Executions
0
Capabilities
–
Artifacts
0
Runtime Failures
0
Network
checking…
Backend Latency
–
Open Tabs
1
Session Uptime
0s
Backend Latency History
Recorded from each "Refresh" click on this tab, kept on this device.
LLM Spend & Circuit Breaker
Estimated LLM provider spend this process's lifetime, and whether the spend-guardrail circuit breaker has tripped -- this is what auto-approval policies check before firing.
Est. spend
$0.00
Total calls
0
Circuit breaker
–
System Health (AuxServices)
Whole-system self-check, separate from the browser-side diagnostics above -- includes whether the tamper-evident audit log (the same one backing the Trading Decision Log) is still cryptographically intact.
Audit log integrity
–
Scheduled tasks
–
Tools registered
–
Backups on disk
–
Revenue Autopilot
One-click control for every job in the Automation Jobs panel below, instead of toggling all 60+ individually. Coverage shows what fraction of registered jobs are currently enabled -- below 100% while Autopilot is on means something drifted (the autopilot_coverage_watch job self-heals this automatically, but you can also just hit Enable again).
Status
–
Coverage
–
Automation Jobs
Every background job Kaizen can run on a schedule -- including the trading safety jobs (drawdown guard, daily loss circuit breaker) that previously had no on/off control anywhere in this UI. Toggling off here stops the schedule; "Run now" fires it once immediately regardless of schedule state.
Loading…
Engine Sessions
Every client (this browser, Terminal, Voice, other devices) that's talked to the engine directly, most recent first.
Click refresh to load.
Engine Diagnostics
Click refresh to run a diagnostic pass against the engine.
Ops Dashboard
Consolidated stats, capability health, connectivity, scheduler/notes counts, process resource usage, and hive status in one view. Owner access required.
Click "Pull dashboard" to fetch the latest snapshot.
Owner-only. Controls every device on the tailnet through the local
tailscale CLI — exit-node routing, subnet routes, connect/disconnect, ping,
and running commands on remote devices (rigs, Arduino-attached hosts, anything with
Tailscale SSH enabled). Device rename and ACL-tag editing aren't exposed by the CLI, so
they're shown read-only here; edit those in the Tailscale admin console. Nicknames below
are local-only labels (see hive.py nickname) and never touch Tailscale itself.Devices
–
Online now
–
Hive peers
–
Device Actions
Actions will show their output here.
Link a New Device (any machine — no tailnet SSH required)
For a device that isn't reachable over Tailscale SSH from here (a fresh machine, one on a
different network, or one you'd rather set up by hand): generate a one-time join command below,
then run it directly on that device's own terminal. No manual secret entry — the command carries
a single-use token that exchanges itself for the real hive credentials.
Requires Kaizen's files to already be on that device (this only links it to the hive — it
doesn't install Kaizen itself there).
Pick a role and press "Generate join command".
Join the Hive not a hive peer
Sends this tailnet device the command to join the Kaizen hive as a compute
peer (over the tailnet, via
tailscale ssh — no manual secret entry needed on your
end). Once joined, the primary node adds it to HiveMindSync's peer list and starts
routing work to it like any other rig.Select a device above, then press Join the Hive.
Device Administration
Approving/removing devices and key-expiry changes require an API token
with tailnet admin scope on the backend; if that scope isn't configured these buttons
will report a clear error rather than silently doing nothing.
Admin actions will show their output here.
Owner-only. Every device currently joined to the Kaizen hive (see
hive.py), whether it's running the full engine or the lightweight
rig_agent.py. Nothing here is a generic remote-command channel — every
action below maps directly to an itemized /hive/run_action category that
each device's own operator has to separately opt into (via
grant-permissions on that device); a device with nothing granted will
refuse with a clear "not granted" error instead of silently doing nothing. Select a
device below to see its status and itemized grants, or check several for a bulk action.Main Instance –
The device the hive treats as its trusted source of truth — its capability/code changes
auto-apply everywhere else instead of being queued for review (
hive.py's
PrimaryNodePolicy). Switching this does not move any running workload by
itself — which device gets new heavy jobs is still decided live by hardware/load
(PowerAwareTaskRouter) — but you'll normally want your strongest box to be both.
Select a device, then press "Make main instance".
Hive peers
–
Online
–
Busy
–
Excluded
–
⚡ Kaizen Safety checking…
Live watchdog on this primary server (see
main.py:
PrimaryCapacityGuard) — separate from any rig's own protection. When this server's own
CPU, RAM, or GPU gets close to its cap, Kaizen refuses to start new heavy dispatch (task/action/
bulk calls) rather than piling on and risking a crash. This is the fix for Kaizen crashing the
server — it now watches its own load, not just remote rigs'.—
🔋 Contribute a device's power to Kaizen
Scans a device's real CPU/RAM/GPU specs and current load, then — if you flip it
on — Kaizen will route work to it up to the safety caps you set below. The caps are enforced
on the device itself (pushed via
rig_agent.py: set_resource_caps), so
it protects itself even if it loses contact with this server. Only use devices you own. This
targets whatever device is currently selected above, or self for this server.No scan yet — pick a device and press "Scan specs".
Kaizen refuses new work on this device the moment
any of these caps is hit — the goal is headroom, not maximum extraction. Lower caps = safer/slower, higher
caps = more throughput/more risk of that device (or this server, for "self") struggling.
Scan a device, adjust caps, then enable.
Selected Device: none selected
"Exclude/Include" is a local routing preference on THIS node only (see
hive.py: PeerExclusionStore) — it doesn't disconnect, notify, or change anything on the
device itself; it just tells this node's own dispatch to stop/resume sending it new work.Select a device above, then choose an action.
Drive/rotate/stop Arduino-class boards (or anything else
classified
microcontroller by hive.py: RigLimbRegistry) directly from
here. Every drive nudge is bounded and self-stopping on the device itself —
see rig_agent.py: _serial_send_timed_drive — so a lost connection, a closed tab,
or a crashed dashboard can never leave a robot driving indefinitely; the worst case is it
coasts to a stop on its own after the nudge duration. The target device's operator still has
to have granted that exact serial port under grant-permissions first. Picking a
microcontroller-classified device above auto-selects it here too.Drive
Full-screen, touch-friendly driving view with the device's own camera as
the background, a choice of on-screen buttons or a thumb joystick, and an optional AI co-pilot.
Select a device and a serial port to begin.
📷 Waiting for camera feed — enable "Live camera background" above, and make sure this device has granted
camera_capture.🐝
Kaizen is currently the one driving this device autonomously.
You'll have full drive control. Tap Exit any time to hand control straight back to Kaizen.
100
speed
Total
0
Success
0
Errors
0
Info
0
🔊 Sound on
🌙 Quiet hours
from
to
(24h)
📶 Connection change popup
Mute rules
Notifications whose title or body contains any of these (case-insensitive) are skipped entirely -- never logged, no sound, no OS popup.
No notifications yet. You'll see task completions, failures, and screen-watch alerts here.
Self-Approval Queue
Requests Kaizen has filed for itself — logins from unrecognized devices, and missions asking for more autonomous steps after hitting their ceiling. Nothing here proceeds until you approve or deny it.
Loading…
Security
Access & Guardrails
Ask before running commands that delete, overwrite, or send irreversible actions.
Safety ceiling on how many chained actions Kaizen can take unattended.
Moving the mouse to a screen corner immediately aborts screen automation (engine-side).
API Key
If your backend requires an API key header, set it here — it's attached to every request as
X-Kaizen-Key.Session Fingerprint
This tab ID–
Backend endpointsame origin
Cross-tab syncon
Logs & Audit
Full Session Log 0
No log entries yet.
🔎
Voice
Used for spoken replies in Chat and Voice tabs.
How fast Kaizen speaks.
Lower is deeper / more human, higher is brighter.
Speak every assistant reply out loud, even outside the Voice tab.
Accessibility
A floating A-/A/A+ control for bumping up the app's text size. Off by default so it doesn't sit in the way; turning it on shows the floating control everywhere (never on the login screen).
Behavior
Allow Kaizen to act without per-step confirmation.
Master toggle for voice features across the app.
Show OS-level notifications for completions and failures.
Broadcast this tab's activity to other open Kaizen tabs.
Current time in the selected time zone (moved here from the top bar).
--:--:--
Same as Voice mode (global) above — kept for legacy wiring.
Same as Autonomous mode above — kept for legacy wiring.
Let this device help with your own questions. Never shared with anyone else, never required.
Off: Kaizen never touches this device. Partial: only when you explicitly ask in that message. Full: Kaizen may also use device control on its own as part of a broader task.
Continuously watches this screen (a screenshot every couple seconds) until turned off. Requires Device Control to be Partial or Full.
Moved here from the top bar, which now only shows Tour.
Show your wandering 3D avatar and companion pet floating on top of the app. Off by default; turn on to bring them onto your screen.
Appearance live preview
Pick any color for the whole UI, or start from a preset. Changes apply instantly and are remembered on this device.
One click recolors both accent colors.
Used for buttons, highlights, and the active tab glow.
#8b5cf6
Used for gradients, badges, and secondary highlights.
#22d3ee
Used for extra highlights, avatars, and mic/voice indicators.
#ec4899
The page's base color, underneath any background image/photo/video.
#070b14
Cards, sidebar, and panel surfaces.
#11182c
Primary body/heading text color.
#f7fbff
Give any single tab in the sidebar its own accent color/tag, independent of the global theme.
Restore default colors, background, and per-tab tags.
Density & Text Size
Fit more on screen, or make everything easier to read. Applies instantly and is remembered on this device.
Comfortable is the default spacing; Compact tightens cards and rows.
Scales all text across the app.
100%
Snappier or more relaxed hover, press, and color transitions across buttons and cards.
100%
Sharper or softer corners on cards, buttons, and inputs.
100%
Size of the Analyze / Trading / Attach / Clear / Export / Clipboard / Emoji row above the message box.
100%
Turns off pulsing badges, streak/typing animations, and other decorative motion.
Background
Upload an image, live photo, or video to use as the app's background — it plays/displays behind the whole UI. Videos and live photos loop silently.
How strongly the background shows through behind the UI panels.
Account
Local profile stored on this device — not synced to any server unless your backend adds real auth.
K
Local / Self-Hosted
Stores your name, email, and avatar on this device.
Removes the locally stored profile from this device.
What Kaizen Should Know & Keep In Mind
This is automatically included with every mission you send from the Chat tab — so Kaizen always has this context, even in a brand-new conversation. (Voice and Terminal send exact commands straight through, unmodified, so engine syntax stays reliable there.)
Connection
All requests go to this base URL. Set to
http://localhost:PORT if you're running the Kaizen server locally — this value is saved across page refreshes.
not tested
Requests never fail on time alone — anything slow is moved to a background job and polled until done. This just sets how often the UI checks in.
Local Model Usage how much of the work Kaizen is doing without calling the LLM
Capability Takeover Ledger native services Kaizen has earned the evidence to handle itself
Learning Gaps what Kaizen currently knows it doesn't know
Self-Knowledge what Kaizen can verify about itself vs. what it only asserts
Data
Chat, tasks, agents, workflows, automations, knowledge base, notifications.
Clears chat, tasks, logs, and metrics from this tab (does not touch the engine's persistent memory).
Bundles theme, appearance, quick notes, streaks, achievements, focus timer stats, and every other on-device preference into one file.
Load a previously downloaded backup file. Overwrites matching settings on this device.
Keyboard Shortcuts
Filter the sidebar navCtrl/Cmd + K
Open command palette (tabs + quick actions)Ctrl/Cmd + Shift + K
Collapse / show sidebarCtrl/Cmd + B
Send messageEnter
New line in composerShift + Enter
Run terminal commandEnter
Cycle terminal history↑ / ↓
Search terminal historyCtrl/Cmd + R
Focus this tab's search box/
Clear/blur a search boxEsc
Open this Help tab?
Jump to a recently visited tabAlt + 1…9
Toggle Zen mode (hide sidebar, dim chrome)Ctrl/Cmd + Shift + Z
Toggle light / dark theme🌙 / ☀️ button, next to the logo
Chat Messages
Edit & resend a message✏️ on any of your messages
Copy one code block📋 Copy button on the block
Read a reply aloud🔊 on any Kaizen message
Mark a reply helpful / not helpful👍 / 👎 on any Kaizen message
Insert an emoji🙂 Emoji button above the composer
Editing a message removes it and everything after it, then resends the edited version through the normal pipeline — same as regenerating, but with your own wording.
Sidebar Extras
Quick Notes scratchpad📝 button, top of sidebar
Focus Timer⏳ Pomodoro button, top of sidebar
Achievements & streaksTrack engagement on the Analytics tab
Chat searchHighlights matches, hides the rest
Jump to pinned messagesButton appears above chat once you pin one
Scroll to latestFloating ↓ button appears once you scroll up
View all pinned messages📋 View pinned button, above chat
Print or save chat as PDF🖨 Print button, above chat
Quote / reply to a message↩️ on any message
Explain what a button does❔ button, top-right of chat — click anything after
Voice Commands
Navigate"open the [tab] section"
Stop speech"stop talking"
Toggle autonomy"enable / disable autonomous mode"
Quick Unit Converter
Handy client-side converter — no network call, works offline.
→
What's New
Frequently Asked Questions
Getting Started Checklist
A quick self-paced checklist if you're new here. Progress is saved locally.
Troubleshooting
Quick Feedback
A faster way to leave a rating and short note than the full issue report below.
Guided Tour
A quick walkthrough of the sidebar and the main areas of the app — handy for onboarding someone new.
Diagnostics & Support
If something's not working right, copy this diagnostic snapshot and include it when you report the issue — it captures your browser, screen, and local app state without any conversation content.
Report an Issue
Describe what happened. This stays on your device — nothing is sent automatically. Use Copy to paste it wherever you file issues.
Activity Timeline
A local, chronological log of what's happened in this session — messages sent, tabs visited, settings changed. Nothing here leaves your device.
Glossary
About this build
Kaizen — your assistant, running here.
No notes yet.
📝
Select a note or create a new one.
Calculator
Habits tracked
0
Done today
0
Longest active streak
0
Today's completion
0%
Focus session
25:00
Cycle 1
Daily goal 0 / 8
0%
Focus minutes today
0
Sessions completed today
0
All-time focus minutes
0
Current daily streak
0
Session Log
Category
From
To
All Units
Recent Conversions
Selected month
$0.00
Today
$0.00
Top category
–
All-time total
$0.00
Avg. daily (selected month)
$0.00
By Category (this month)
All Expenses
Generator
Strength Checker
Enter a password above to see its strength.
Recently Generated (session only, not saved)
🎲 Dice Roller
d
–
🪙 Coin Flip
–
🎡 Random Picker
🔢 Random Number
to
–
Words: 0
Characters: 0
Characters (no spaces): 0
Sentences: 0
Paragraphs: 0
Reading time: 0 min
Transform
Find & Replace
Stopwatch
00:00.00
Timers
Today's Entry
Mood — Last 30 Days 🔥 0-day streak
Past Entries
📋 To Do 0
🚧 In Progress 0
✅ Done 0
Progress0 / 0
Tip percentage: 15%
Tip amount
$0.00
Total with tip
$0.00
Per person
$0.00
Ready
Cycles completed: 0
Total capabilities tracked
–
Critical (fail rate > 30%)
–
Hot (high demand)
–
Stale (> 7 days since fix)
–
Top Capabilities by Priority Score 7-day window
Composite = demand × failure_rate × time_since_fix. Engine evolves these first.
Click Refresh to load heatmap data from the engine.
Raw Heatmap Output
Waiting for refresh…
Winning Mutation Diffs 0 stored
Every gate-passing, approved diff stored verbatim. Used as few-shot examples when generating new mutations — semantically similar past winners are injected into the generation prompt automatically.
Click Refresh to load replay buffer.
Model health
–
Classifier accuracy
–
Classes in table
–
Episodic memories
–
Training steps
–
Loss trend
–
Confusion Leaders Worst-accuracy classes
Classes the model most often confuses on the held-out validation set. These are candidates for auto-merge or targeted re-training.
Click Refresh to load.
Curiosity Queue 0 items
Low-confidence requests queued for the idle self-study loop — targeted gaps the model knows it doesn't know yet. Drains automatically during idle cycles.
No items in curiosity queue.
Dataset Ingestion
Queue a public dataset URL (plain-text or JSONL) for background ingestion. Rate-limited and deduplicated — restarting the process won't re-train the same data twice.
Raw Health Report
Click Refresh to pull local_model_health from the engine.
Injection incidents (all time)
–
Injection incidents (24h)
–
Drift warnings (7d)
–
Open flaws detected
–
Prompt Injection Incidents 0
Every injection detected by the 3-layer scanner (pattern matching, structural anomaly, context rupture). Raw request is sanitized; sanitized version reaches routing, never the original. Detection sensitivity configurable via
KAIZEN_INJECTION_SENSITIVITY.No injection incidents on record. Click Refresh to check.
Behavior Drift Alerts 0
BehaviorFingerprintEngine compares each capability's output-hash vector after every hot-reload. Cosine distance > 0.15 from baseline blocks auto-approve and fires a BEHAVIORAL_DRIFT event, even if CodeValidator passes.
No drift alerts. Click Refresh to check.
Flaw Detector Report 0 open
FlawDetector combines unsupported request patterns, high-fail-rate capabilities, and semantic gap analysis against known "useful AI agent primitives". All three signal streams merge into a ranked flaw report queued into the improvement backlog automatically.
Click Refresh to load flaw report.
Raw Security Report
Waiting for refresh…
Tokens this cycle
–
Cycle budget
–
Budget used
–
Calls deferred
–
Calls truncated
–
Calls downgraded
–
Budget Progress
Loading…
Per-Model Usage Breakdown
Click Refresh to load usage breakdown.
Budget Governor Config
Adjust budget limits live — takes effect on the next evolution cycle. Env vars:
KAIZEN_LLM_BUDGET_TOKENS_PER_CYCLE, KAIZEN_REPLAY_BUFFER_SIZE.Raw Token Budget Status
Click Refresh to pull token_budget_status from the engine.
Total traces stored
–
Most common failure gate
–
Avg attempts per run
–
Failure Gate Breakdown
Click Refresh to load.
Recent Causal Traces 0
Each trace shows the objective → attempt chain → root cause. Stored in KB under
causal_trace_v1; fed back into the NEXT mutation for the same objective domain.No causal traces loaded. Click Refresh.
Raw Causal Trace Report
Click Refresh to pull causal_trace from the engine.
Current interval (s)
–
Aggression (0–1)
–
Mode
–
Recent success rate
–
Recent fail rate
–
Resource headroom
–
Aggression Timeline
How boldly mutations are being proposed over recent cycles — higher = more aggressive diffs, broader objectives, larger candidate pools.
–
Tune Scheduler Knobs
These map directly to
KaizenConfig fields. Changes take effect on the next scheduling tick.Raw Scheduler Status
Click Refresh.
Sync repo
–
Last sync
–
PRs pushed (this session)
–
Patches pulled & applied
–
Recent Auto-Improvement PRs 0
Each row is a draft PR opened by Kaizen after a successful self-mutation. Click the link to review or merge on GitHub.
Click Refresh to load.
Configuration
Set via env vars on the server — these fields are read-only here. To enable: set
KAIZEN_SYNC_REPO=owner/repo and KAIZEN_GITHUB_TOKEN.Click Refresh to show current config.
Raw GitHub Sync Status
Click Refresh.
Internet reachable
–
Latency (ms)
–
Last probed
–
Capabilities Available Offline
These run as plain local Python with no LLM call — they keep working even when there's no internet and no model API is reachable.
Click Refresh to load.
Provider Reachability
Which LLM providers and search backends are currently reachable from this machine.
Click Refresh to load.
Raw Offline Status
Click Refresh.
Cached completions
–
Cache hits (session)
–
TTL window
–
LLM calls saved
–
Recent Cache Entries 0
Most recently cached completions, newest first. Mutation requests bypass the guard by default.
Click Refresh to load.
Raw Idempotency Status
Click Refresh.
Total entries
–
Log size
–
Event types
–
Event Type Breakdown
Click "Load Stats" to populate.
Search Audit Log
Search results appear here.
Total packages installed
–
Auto-install enabled
–
Most recent install
–
Package Registry 0
Click Refresh to load installed packages.
Safety & Vetting Info
Every package here was approved by PackageSafetyChecker before installation, then sandbox-imported in a subprocess to verify it loads cleanly. Auto-install is opt-in via KAIZEN_AUTO_INSTALL_PACKAGES=1. To use a venv instead of the active Python environment, set KAIZEN_VENV_INSTALL=1.
Active plans
–
Completed plans
–
Failed plans
–
Sub-goals total
–
In-Flight Plans 0
Plans currently executing. Each sub-goal shows its status independently —
stop-on-first-failure or best-effort-continue is set per-plan at decomposition time.No active plans. Click Refresh.
Plan History 0
Click Refresh to load plan history.
Planner Config
Failure propagation policy and sub-goal timeout are set via
KaizenConfig. Multi-step requests are detected automatically by a lightweight heuristic (conjunctions + verb phrases + entity count).Click Refresh to show planner config.
Raw Planner Status
Click Refresh.
Content & Social
Connect accounts with an API access token (never a password -- Kaizen never logs in as you). Every draft sits pending until you approve it here, unless you've turned autopilot on for that account.Draft a post
Review queue 0
+ Connect a new account
Connected accounts 0
Channel growth real posted-content data
Total content income across every connected account
Sponsorships 0 pending
+ Log an inbound sponsorship offer
Earnings ledger 0 pending
Withdrawals 0 pending
+ Request a withdrawal
Trends & sponsor leads discovery, not auto-action
Pipeline performance 30d
Best-performing platform by $ earned
Content Library 0
Caption Tools
Templates 0
Pending Review 0
Scheduled Posts 0
RSS Watches 0
Keyword Watches 0
Web Browser
Kaizen's own independent stdlib browser -- no Playwright/Selenium/DDG/Brave dependency. Direct URLs fetch straight; anything else is smart-routed through the SearxNG/DDG/Brave research chain with a key-free fallback if every backend is down.Saved logins 0
History
Social Browser
Quick-launch real browser tabs, or open a live scripted session. You always log in yourself in a visible window -- Kaizen never stores or replays passwords, tokens, or session cookies unless you explicitly ask it to save one (see "Log in" per session below). Live sessions only do what you ask, one action at a time.Setup status
Quick-launch
Live sessions 0
Action console
Screenshot gallery
Session address book 0