Models, Skills & Limits
Mahout routes your prompt through a model and a set of available research skills.
Models
The model picker shows the models currently available to your account, and you can set model preferences so Mahout uses your chosen model by default. Availability can change as Airavat evaluates cost, latency, quality, and provider reliability.
Use the default model when you are unsure. Change models when you need a different speed, depth, or reasoning profile.
Effort levels
Effort controls how much thinking and output budget Mahout spends on a prompt. The available options depend on the selected model and your account.
| Effort | What it does |
|---|---|
| Low | Smallest thinking and output budget. Fastest. Best for quick summaries, simple explanations, and lightweight follow-ups. |
| Medium | A moderate thinking and output budget. Good for normal research, comparisons, and source-backed answers. |
| High | A larger budget for longer multi-step synthesis and code-backed work. |
| X-High and Max | Raise the thinking and output budget further for the longest, most involved tasks. |
Higher levels mainly raise the budget rather than switching to a categorically deeper reasoning mode, so treat them as a longer budget, not a distinct deeper tier. Higher effort can be slower and use more of your usage budget. If you do not see X-High or Max, the selected model or your account does not expose those levels.
Skills
Skills are model-accessible tools that let Mahout gather data, run analysis, or produce structured outputs.
Available skills include:
- asset and market analysis over historical bars (Databento) and centralized-exchange history (Binance, Bybit, OKX)
- crypto market data (CoinGecko)
- macro data (FRED, US Treasury, macro-economic calendar)
- SEC EDGAR filings research
- GDELT news search and fetching a URL you paste
- sandbox-backed code execution for supported analysis
- Polymarket trading data (top traders, trending markets, market detail, search, and your own fills and positions)
- Polymarket filter creation (via chat contract
pm_filter_create) - Polymarket filter coverage estimates (via chat contract
pm_filter_estimate) - backtests
- artifact generation and reruns
- agent drafting, compilation, historical preview, deployment, and fire analysis
The visible skill list may be simplified compared with the internal routing Mahout uses.
Rate and usage limits
Mahout has usage controls to keep the product stable.
You may hit a limit if:
- too many requests are sent quickly
- too many streams are active
- a model or sandbox budget is exhausted
- a provider is degraded
When a limit is hit, wait and retry later. If the issue persists, contact support with the approximate time, session, and what you were trying to do.
Degraded states
Mahout can continue in a reduced mode when some dependencies are unavailable.
Examples:
| Degraded area | Expected behavior |
|---|---|
| Model provider outage | Mahout may show an unavailable or fallback response. |
| Data source outage | Mahout should explain the missing data source and use other sources if possible. |
| Sandbox outage | Chat may still work, but code-backed artifacts or reruns may fail. |
| Session problem | You may need to refresh or log in again. |
If a response looks incomplete, ask Mahout what failed or which sources were unavailable.
Best practices
- Use low or medium effort while exploring.
- Switch to high effort for final synthesis or backtest interpretation.
- Ask for sources on important answers.
- Save final outputs as artifacts.
- Preview an agent against historical time before deploying it.
- Start a new session when the topic changes materially.