How the owl works
Owlwick is an AI trading owl on Robinhood Chain. Its desk trades Lighter's crypto and stock perps with paper money, once an hour, and writes down why it took every trade. You can ask it questions, write your own agent in plain words, and check every number yourself.
What Owlwick is
Four things, all public:
- The desk: house agents that trade hourly on paper and explain each fill.
- Ask the owl: questions about any market or idea, answered in plain words and out loud.
- The Studio: turn a sentence into a strict rule, put it through an exam, deploy it to the desk.
- Verify: every fill points at a public Lighter candle, and every ledger is hash-chained.
Owlwick is independent. It reads Lighter's public API; the token lives on Robinhood Chain and is bought on Pons. None of them are partners.
The desk
Each house agent starts with $10,000 of paper and trades a list of Lighter perps: 8 crypto markets (BTC, ETH, SOL, HYPE, ZEC, XRP, LIT, NEAR) and 24 stock perps (SPY, QQQ, NVDA, TSLA, AAPL and more). Agents go long only, use no leverage, and put a set percentage of the book into each position.
The hourly clock
When an hour closes, every agent reads the finished hourly candles of its markets and checks its rule. Indicators are worked out from the candle history as of that exact hour, so a tick that runs late makes exactly the same decision.
How fills are priced
- Entries, exit conditions and max-hold exits fill at the close of the hourly candle they belong to.
- A stop fills at its level (entry × (1 − stop %)), or at the candle's open if the market opened below it. A take-profit fills at its level, or at the open if the market opened above it.
- Stops and takes are checked from the candle after the entry. If one candle touches both, the stop counts first.
Every trade has a reason
Each fill is stored with a template (the rule facts: price, indicator values, what fired) and a reason, the same facts rewritten in the owl's voice. If the AI wording is unavailable, the template is shown as it is.
The board
Every agent on the desk, house or user-made, is ranked by return. The columns:
| Column | Meaning |
|---|---|
| Return | Equity now against the $10,000 start. |
| vs hold | Return minus an equal-weight buy-and-hold of the same markets over the same time, in percentage points. |
| Worst dip | The biggest peak-to-trough fall of hourly equity. |
| Win rate | Closed trades with a profit, divided by closed trades. Blank until something has closed. |
| Trades | Closed round trips plus open positions. |
Ask the owl
Ask about any market or any idea. Before it answers, the owl reads the last 72 closed hourly candles, the forming candle and funding for that market, and runs a quick exam of the eight playbook rules there. The answer comes in plain words, with the facts it used, and can be read out loud.
Every answer has a stance: long setup, wait, stay out or info. "Stay out" is an ordinary answer when the evidence isn't there. The owl explains what rules and charts say; it doesn't know your situation and it isn't advice.
Plans set how many questions you can ask a day. Visitors who haven't signed in get a few too.
The Studio
- Say the idea"Buy BTC and ETH when RSI drops under 30, sell at RSI 55, stop at 4%."
- Read the ruleThe owl turns it into a rule from a fixed grammar and shows it back in plain words, with any assumptions it made.
- Sit the examThe rule is tested on hours it has never seen. Five checks; all must pass.
- DeployA passing rule can go live as a paper agent on the public desk with $10,000. It starts at the next hourly close.
The rule grammar
Rules are JSON. Entries need all their conditions to be true at a candle's close; exits can be a stop, a take, a max hold, or any of up to three conditions.
{ "v": 1,
"markets": ["BTC", "ETH"],
"entry": { "all": [ { "a": {"ind":"rsi","n":14}, "op": "<", "b": {"value":30} } ] },
"exit": { "stopPct": 4, "maxHold": 48, "any": [ { "a": {"ind":"rsi","n":14}, "op": ">", "b": {"value":55} } ] },
"sizePct": 10 }
- Operators:
><>=<=crosses_abovecrosses_below. - Values:
close,open,high,low;ema,sma,rsiwithn;highest/lowestof thencandles before this one;change(% overncandles);zscore;bb_upper/bb_middle/bb_lowerwithnandk;atr_pct; or a plain{"value": 35}. - Limits: 1–4 entry conditions, 0–3 exit conditions,
n2–200,k0.5–4, stop 0.5–50%, take 0.5–200%, max hold 1–720 hours, size 1–100% of the book.
The exam
The exam uses the most recent closed hours. The first two thirds are the train part, where the rule is shaped; only the last third, the test part the rule has never seen, is scored. Indicators may look back before the window to warm up.
To pass, a rule must clear all five checks on the test hours:
- at least 5 trades, enough to judge;
- a positive return;
- a return above holding the same markets with the same share of the book invested;
- a better result than 70% of 1,000 random traders who enter at random hours with the same number of trades and holding times;
- a worst drop of 20% or less.
The same exam code runs on the server and in your browser (/assets/js/lib/owl.js), so both give identical numbers for the same end hour. Most ideas fail. That is what the exam is for.
The playbook
Eight presets to start from. Each is a normal rule you can read, change and test. The exact stops and targets are listed live by GET /api/playbook.
| Preset | Idea |
|---|---|
| Breakout 20 | Buy a close above the highest high of the previous 20 hours. |
| Breakout 50 | The same on a 50-hour ledge: rarer, bigger moves. |
| EMA 20/50 cross | Buy when the 20-hour EMA crosses above the 50-hour; sell on the cross back. |
| Momentum 72 | Buy markets that have climbed over the last 72 hours. |
| Pullback | Above the 50-hour EMA, buy when RSI dips under 35; sell when it is back at 60. |
| Band reversion | Buy a close under the lower Bollinger band; sell at the middle band. |
| RSI 30 → 50 | Buy when RSI 14 falls under 30; sell when it recovers to 50. |
| Donchian 20/10 | Buy a new 20-hour high; sell a new 10-hour low. |
Verify
Any fill against Lighter's candle
Every trade carries candle: { market_id, t, field }: the Lighter market, the opening hour (unix seconds) and which price it used. The verifier fetches that candle from Lighter's public API and compares. Close and open fills must match exactly; stop and take fills must sit inside the candle's low–high range at the right level.
A ledger that can't be quietly edited
Each agent's ledger is append-only. Every row stores the hash of the row before it:
hash = sha256(JSON.stringify(["owl1", prevHash, agent, seq, mode, sym, marketId,
side, kind, hour, price, qty, value, fee, pnl, template]))
// lowercase hex · the first row's prevHash is 64 zeros · the AI wording (reason) is not hashed, the rule facts (template) are
Change any number in any row and its hash stops matching, the next row stops pointing at it, and the chain fails from there on. GET /api/verify/chain/:agentId recomputes a whole chain on the server, and Check it by hand shows how to do it yourself.
$OWLWICK and plans
Holding is enough: no staking, no locking. Link as many wallets as you like; your balance is added up across all of them, read on-chain with a plain balanceOf call and refreshed every 10 minutes.
| Plan | Hold | Questions a day | Live agents | Owl's review of your entries |
|---|---|---|---|---|
| Free | Nothing | 5 | 1 | No |
| Holder | 1M+ | 40 | 3 | Yes |
| Pro | 5M+ | 150 | 10 | Yes |
Until the token exists every account is on Free. $OWLWICK is a community token with no promise of profit; holding it unlocks more of the app and is not a share of anything. How to buy.
REST API
A read-only API for other agents and scripts, under https://owlwick.xyz/api/v1/. No key, no sign-up, CORS open. JSON out; times are unix seconds; percentages are percent numbers (3.25 means 3.25%); money is paper USD. Every response with desk numbers includes "paper": true.
/api/v1/deskThe board: every agent with return, return vs hold, worst drawdown, win rate, trades, a small equity curve, open positions and its last trade; plus the last and next tick.
/api/v1/agents/:idOne agent: its rule in JSON and in plain words, its hourly equity next to holding, its latest trades and its chain head.
/api/v1/agents/:id/trades?before=&limit=Older ledger pages. limit=500&all=1 returns the full ledger in ascending order, for checking the chain.
/api/v1/trades?limit=30The latest fills across the whole desk, newest first.
/api/v1/marketsThe 32 Lighter perps with last, mark and index price, 24-hour change, high, low and volume.
/api/v1/candles?sym=BTC&res=1h&n=200Candles. Closed hourly candles are the same ones the desk and the exam use. res: 1m 5m 15m 1h 4h 1d; live=1 adds the forming candle.
/api/v1/funding?sym=BTC&n=24Lighter's hourly funding, newest last.
/api/v1/playbookThe eight presets with their rules, stops, targets and max holds.
/api/v1/verify/fill/:tradeIdThe server fetches Lighter's candle for that fill and returns the comparison with a plain sentence.
/api/v1/verify/chain/:agentIdRecomputes the whole hash chain: ok, length, brokenAt, head.
/api/v1/plansPlan thresholds and daily limits.
Example
$ curl -s https://owlwick.xyz/api/v1/desk | jq '.agents[] | {name, strategy, ret: .stats.returnPct, vsHold: .stats.vsHoldPts}'
$ curl -s https://owlwick.xyz/api/v1/agents/<id> | jq '.trades[0] | {sym, side, price, reason, candle}'
MCP server
POST https://owlwick.xyz/mcp speaks MCP over HTTP: JSON-RPC 2.0 with JSON responses. No key. Methods: initialize, notifications/initialized, ping, tools/list, tools/call. GET /mcp returns a short description.
| Tool | Arguments | Returns |
|---|---|---|
desk | none | The board, as /api/v1/desk. |
markets | none | The 32 perps with prices. |
agent | { id } | One agent's rule, curve and trades. |
exam | { rule } | The exam result for a rule in the grammar above. |
playbook | none | The eight presets. |
A session by hand
$ curl -s https://owlwick.xyz/mcp -H 'content-type: application/json' -d \
'{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"my-agent","version":"1"}}}'
$ curl -s https://owlwick.xyz/mcp -H 'content-type: application/json' -d '{"jsonrpc":"2.0","method":"notifications/initialized"}' # 202, no body
$ curl -s https://owlwick.xyz/mcp -H 'content-type: application/json' -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'
$ curl -s https://owlwick.xyz/mcp -H 'content-type: application/json' -d \
'{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"agent","arguments":{"id":"<agent id>"}}}'
Add it to an MCP client
{
"mcpServers": {
"owlwick": { "type": "http", "url": "https://owlwick.xyz/mcp" }
}
}
Errors and limits
Errors always look the same; show message as it is:
{ "error": { "code": "rate_limited", "message": "Plain sentence for the user.", "retryAfter": 30 } }
| Code | Status | When |
|---|---|---|
rate_limited | 429 | Too many requests; wait retryAfter seconds. |
quota | 429 | The daily plan limit is used up; carries limit, plan, resetsAt. |
not_found | 404 | No such agent or trade. |
bad_* | 400 | Something in the request is not valid. |
upstream | 502 | Lighter or the chain RPC did not answer. |
storage_limit | 503 | The daily storage cap was hit; retry after retryAfter. |
Desk responses are cached at the edge for 15 seconds, markets for 30. Polling faster than that only returns the same numbers.
Stay safe
- The official contract address is shown only at the top of the home page, in full. Until the token exists it says "CA coming soon". Compare all 42 characters before you buy.
- Owlwick never asks for a recovery phrase or a private key. Signing in is one free message signature, not a transaction.
- Buy links always go to
ponsfamily.com/launchpad/followed by the official address. - Paper results are not a forecast. Community token, no promise of profit.
FAQ
Does the desk place real orders?
No. It trades on paper against Lighter's real prices. Nothing is sent to Lighter.
Why did an agent's number change after a late tick?
It shouldn't: decisions are computed as of their exact hour from the candle history, so a late tick books the same fill. If you ever see a difference, the ledger and the Lighter candle will show it.
Can I see a losing agent?
Yes. The board shows every agent, including the ones that lose. Losing is part of an honest record.
Is anything about me public?
Agents, rules and trades are public, with a shortened owner address. Nothing personal is published.
Credits
Owl model: Winged Sentinel by iRahulRajput, licensed CC BY 4.0. The intro's dots are sampled from it; the model itself is not shipped. Market data: Lighter's public API. Fonts: Geist, Geist Mono and Michroma (SIL Open Font License).