AI market research (multi-agent analysis)
Run an advisory multi-agent research debate on a supported crypto pair with your own AI provider key, read the agent reports, and understand the daily allowance.
Availability: AI research is available on paid plans. It appears in your dashboard only once it has been enabled for your account. Until you see it there, nothing on this page is something your account can use today. The dashboard is the source of truth for what is available to you.
AI research is advisory only: it is generated text, not financial advice, and it never places a trade.
What AI research does
You choose a supported crypto pair and an AI provider you already store a key for. A team of AI agents then researches that pair and debates a conclusion:
| Agent | What it contributes |
|---|---|
| Market analyst | Price action and technical picture. |
| Sentiment analyst | Social and community sentiment. |
| News analyst | Recent news flow. |
| Fundamentals analyst | Project and macro fundamentals. |
| Risk judge | The verdict of the risk debate. |
| Trader agent | The proposed plan the debate converged on. |
When the run finishes you get one written report per agent plus a single advisory label — Buy bias, Hold bias, or Sell bias.
A run usually takes several minutes and is stopped automatically if it passes roughly twenty minutes. You can cancel a running analysis at any time.
What it never does
- It never creates, configures, starts, or stops a bot.
- It never places, cancels, or sizes an order.
- No bot ever reads an analysis or acts on its advisory label.
The label is a summary of what the agents argued, written for a person to read. It is not a recommendation, not a signal feed, and not investment, legal, or tax advice. Analyses can be incomplete or simply wrong.
You bring your own AI key
AI research runs on your provider key, so inference is billed to you by that provider and never marked up by the platform. Add a key under AI settings first; only providers you have stored a key for can be selected.
| Provider | Financial memory |
|---|---|
| OpenAI | Available |
| Anthropic | Not available |
| Not available | |
| DeepSeek | Not available |
| Kimi | Not available |
| GLM (Z.ai) | Not available |
Financial memory lets a run recall the conclusions of your earlier analyses. It needs both language-model and embedding access on one key, and OpenAI is the only provider whose stored key gives us both. On every other provider the memory steps are skipped, the run still completes, and the analysis is labelled Financial memory is off for this provider.
You pick two models per run from the models your provider supports: a deep reasoning model for the slower, higher-quality steps and a fast model for the shorter research steps. Both must come from the same provider.
Degraded data sources
Some market data sources need their own external access. When one is unavailable, the run does not fail — it continues without that source and lists what was missing, so you can weigh the result accordingly. A run that lists degraded sources is a thinner run, not a broken one.
Supported pairs
Research is limited to a curated crypto universe, priced in USD:
ADA-USD, AVAX-USD, BCH-USD, BNB-USD, BTC-USD, DOGE-USD, DOT-USD, ETH-USD, LINK-USD, LTC-USD, SOL-USD, XRP-USD
A pair outside this list is refused. The list is curated deliberately — it is not a search box.
Daily allowance
Each analysis you start counts against a per-plan daily allowance. The window is a UTC day, so the allowance resets at midnight UTC, not at your local midnight.
| Plan | Analyses per UTC day |
|---|---|
| No paid plan | 0 |
| Starter | 3 |
| Pro | 10 |
| Enterprise | 50 |
Cancelled, failed, and timed-out runs still count — the allowance limits how many runs you start, not how many succeed. The dashboard shows how much of today's allowance you have used.
How to run an analysis
- Open AI research in the dashboard.
- Pick a pair from the supported list.
- Pick a provider you have stored a key for, then a deep reasoning model and a fast model.
- Start the analysis. It appears immediately with a progress indicator; you can leave the page and come back.
- When it finishes, open it from History to read the agent reports and the advisory label.
How to read the result
- Status — queued, running, completed, failed, cancelled, or timed out.
- Advisory label — buy / hold / sell bias, present only on a completed run.
- Agent reports — one section per agent, shown as plain text.
- Notes — whether financial memory was off, and which data sources were degraded.
Read the reports, not just the label. The label compresses a long debate into one word and loses every caveat that made the debate worth reading.
History and retention
Analyses are listed newest first and are yours alone — no other account can see them. Runs are kept for 90 days and then deleted automatically. Deleting your account deletes them immediately.
Automation
The same three research actions are available to the in-app AI assistant and to external MCP clients. See AI research tools for chat and MCP.