Mahout
Mahout is Airavat’s AI copilot. It lives inside the Airavat dashboard and helps you research assets, macro themes, filings, and news; run analysis and backtests; save outputs as artifacts; and deploy agents that watch the market for you — all in one chat-based workspace.
What Mahout is for
Use Mahout when you want to:
- explore an asset, macro theme, filing, or event before deciding what to research next
- ask follow-up questions without rebuilding the same context each time
- run backtests and code-backed analysis — AI-generated code executes in an isolated sandbox
- generate charts, tables, and research notes, and save them as artifacts you can return to and rerun
- turn a rule you care about into an agent that watches the market and notifies you when it fires
Agents
Agents are a first-class part of Mahout. You describe a rule in plain language, and Mahout drafts the agent, compiles the rule into a machine-executable form, and can preview it against historical time before you deploy it. A deployed agent runs on a schedule, watches its data sources, and sends you an in-app notification each time it fires. You can pause, resume, or archive agents at any time.
Agents notify; they do not trade. A rule that asks for order execution is rejected at compile time. See Research Workflows for the full agent workflow.
What Mahout is not
Mahout is not an execution product. It does not place trades, move funds, or operate accounts for you — and neither do its agents.
It can help with research and analysis, but you remain responsible for decisions, sizing, execution, and risk management.
Where to find it
If your account has access, open Mahout in the Airavat dashboard.
If you do not see Mahout in the dashboard sidebar, your account has not been enabled yet.
How access works
Mahout has its own product permission, separate from other Airavat products. If you do not see Mahout in the dashboard sidebar, your account has not been enabled for it yet.
Data sources
The copilot can draw on:
- Databento historical bars
- Binance, Bybit, and OKX centralized-exchange history
- CoinGecko crypto market data
- FRED (Federal Reserve Economic Data)
- US Treasury fiscal data
- SEC EDGAR filings
- GDELT news
- a macro-economic calendar with point-in-time reconstruction
- Polymarket trading data (top traders, trending markets, market detail, search, and your own fills and positions)
For web research, Mahout can search GDELT news and fetch a URL you paste into the chat. There is no general web search.
Main areas
- Getting Started: first session, first prompt, model controls, and basic navigation
- Research Workflows: asset and macro research, backtests, agents, and follow-up prompts
- History: reopen previous sessions, understand session context, and keep research threads separate
- Artifacts: charts, tables, reports, saved research, and reruns
- Data & Permissions: source availability, provenance, freshness, and permission behavior
- Models, Skills & Limits: model choice, effort levels, skills, usage limits, and degraded states
- Troubleshooting: common problems and what to do
Working style
Mahout works best when you give it a concrete research target and the decision you are trying to support.
Good prompts:
- “Analyze BTC over the last 90 days. Focus on momentum, volatility, and key macro catalysts.”
- “What did US inflation data do over the last year, and what is on the macro calendar next week?”
- “Build a simple mean-reversion backtest for ETH hourly candles from 2022 through 2024.”
- “Summarize what changed in this market since yesterday and list the data sources used.”
Avoid prompts that ask Mahout to trade for you or treat model output as guaranteed truth.