Explainable crypto research
See the argument behind every investment decision.
Andur convenes a committee of specialized AI agents, records where they agree and disagree, applies a binding risk gate, and gives you the evidence to make the final call.
Research and paper trading first. No promised returns. No hidden signal.
Free account · Simulated funds · Live trading remains off unless explicitly enabled
Live
Crypto market prices
6
Independent agent votes
1
Binding risk veto
Paper
Default execution mode
Decision method
A repeatable process, not a magic prompt
Each analysis follows the same sequence. The model contributes reasoning, while scoring, risk enforcement, supervision, and recordkeeping remain explicit system steps.
01
Gather the same evidence
Every agent receives the same market snapshot, candles, recalled context, news, and sentiment inputs.
02
Record independent views
Six intelligence agents return a stance, conviction, rationale, and score contribution on the record.
03
Apply hard risk limits
Liquidity, volatility, sizing, concentration, and exposure rules can block a trade regardless of the AI's opinion.
04
Score and adjudicate
A deterministic scorecard produces the action and confidence. The CIO model writes the plain-language thesis.
05
Persist the full record
The recommendation, votes, debate, approvals, and risk checks are stored so the decision can be inspected later.
06
Grade what happened
After an outcome matures, agent calls are compared with the realized move and future vote weights are adjusted.
What you receive
Inspect the decision before acting
A recommendation is not just a buy or sell label. Andur preserves the supporting record so you can challenge the conclusion—or ignore it entirely.
Action and confidence derived from the score
Bullish, bearish, and neutral votes on one axis
A factor-by-factor score breakdown
Written rationale from each participating agent
A binding pass or block from the risk engine
A timestamped record for later evaluation
Risk outranks the model
The risk engine can veto an order even when every intelligence agent is bullish.
Limits are evaluated in application code and checked again before execution. They are not suggestions embedded in a prompt, and an approval cannot bypass a block.
Human supervision
Recommendations can require explicit approval. Paper execution is the default; connecting a real venue is a separate, deliberate action.
Product disclosure
Clear about what is real today
Credible research software should make its boundaries as visible as its capabilities.
What is live
Crypto prices, asset fundamentals, market sentiment, and news headlines all come from live sources. Committee orchestration, scoring, risk checks, the systematic strategy ensemble, backtesting, persistence, and LLM-authored theses all run in the application.
What is simulated
Paper trading is the default execution venue. The sentiment reading is the market-wide Fear & Greed index, so it carries no per-asset social measurement and reports none. Fundamentals cover no on-chain data — holder distribution, exchange flows, and protocol revenue need a dedicated provider, and every report says which factors it could not measure.
What we do not claim
Andur does not promise returns or present simulated results as proof of future performance. Backtests fill at the next bar's open and charge fees and slippage on every trade, but a result on historical bars is still not evidence of future returns. This is research software, not financial advice.
Inspect the current agent record
The performance endpoint reports graded sample sizes and agent weighting without requiring an account.
Use AI as an accountable analyst, not an oracle.
Start with live market research and simulated capital. Review every vote, challenge the thesis, and decide for yourself.
Create a free account