Aurora AI Agent
How to use Aurora for strategy creation, research, and portfolio management.
Aurora AI Agent
Aurora is the world's only publicly-available, truly autonomous AI agent for algorithmic trading. She can create portfolios, build strategies, run backtests, optimize parameters, screen stocks, analyze earnings, generate research reports, and deploy strategies to live markets — all through natural conversation, and all without writing a single line of code.
Aurora goes far beyond a simple chatbot. Powered by the ReAct framework (Reasoning and Acting), Aurora iteratively reasons through complex financial problems, takes actions, observes the results, and adjusts her approach — fully autonomously.
How Aurora Decides What To Do
There's a single place to talk to Aurora — open Aurora. You don't pick a mode. Every message you send is automatically classified by Aurora's router and handled one of three ways:
- Direct answer. For requests that resolve in one step — create a portfolio, answer a research question, run a single backtest — Aurora just answers, in the same conversation.
- Multi-step agent run. When a goal needs several dependent steps (research → create → backtest → compare → optimize), or refers to something that has to be interpreted before Aurora can act on it, Aurora automatically escalates: it writes a numbered plan and executes it autonomously using the ReAct loop, right in that same conversation.
- Clarifying question. If the request is ambiguous, Aurora asks a follow-up instead of guessing.
You never have to decide which mode to use — Aurora reads the request and picks the lightest approach that fully satisfies it.
Example prompts that get a direct answer:
"Create a portfolio that dollar-cost averages into the S&P 500""Which tech stocks have a PE ratio below 20 and positive free cash flow?""Analyze NVIDIA's latest financial statements""Backtest my portfolio from 2020 to today""Build an SPY bull put spread when price is above the 200-day SMA, 30–45 DTE, 10% of buying power" (options — see Options trading)
Aurora often provides suggested follow-up prompts to help you refine your ideas.
Example prompts that trigger a multi-step agent run:
"Create a strategy that rebalances between UPRO and GLD. Figure out the best allocation split and rebalancing frequency.""Research oil companies that benefit from Venezuelan sanctions relief, create a portfolio, backtest it, and optimize it""Find the best dividend stocks, build a diversified portfolio, compare it against SPY, and optimize the parameters"
When Aurora escalates into a multi-step run, it works through the ReAct loop:
- Plan: Aurora breaks your request into a sequence of steps.
- Think: She reasons about what to do next based on what she's observed so far.
- Act: She calls one of her tools (create a portfolio, run a backtest, screen stocks, etc.).
- Observe: She analyzes the results.
- Repeat: Steps 2-4 repeat until the task is complete.
Aurora can create dozens of portfolio variants, backtest them all, compare results, and deliver an optimized solution — all without you having to intervene.
Automation Settings
Once Aurora escalates into a multi-step run, it can operate at two levels of autonomy. This is a setting, not a mode you navigate to:
- Automated (default): Aurora executes her plan fully autonomously. She plans, acts, and delivers results without needing approval at each step.
- Semi-Automated: Aurora pauses after generating her plan and waits for your approval before executing. Useful when you want to review and refine the approach before she starts.
You configure this from the chat/agent settings.
Aurora's Tools
Aurora has access to a wide range of specialized tools:
| Tool |
What It Does |
| Create Portfolios |
Build complete portfolios with strategies, conditions, and indicators from a natural language description |
| Backtest Portfolios |
Run backtests on your portfolios over any historical date range |
| Optimize Portfolio |
Use genetic algorithms to fine-tune strategy parameters |
| AI Stock Screener |
Screen stocks using fundamental and technical criteria with AI-powered filtering |
| Multi Stock Screener |
Screen across multiple criteria simultaneously |
| Analyze Earnings |
Deep analysis of a company's earnings reports and financial statements |
| Deep Research |
Generate comprehensive research reports on stocks or topics |
| Create Watchlist |
Set up stock watchlists for monitoring |
| Read Backtest |
Analyze and interpret backtest results |
| Read Optimization |
Analyze and interpret optimization results |
| Knowledge Search |
Answer general questions about trading, finance, and NexusTrade |
| Stock News |
Search and analyze real-time stock news |
| Public Portfolios |
Search and analyze other users' shared portfolios |
| Fetch User Portfolios |
Retrieve your deployed portfolios with performance statistics and strategy details |
| Fetch User Watchlists |
List your watchlists and the symbols in each |
| Edit Watchlist |
Rename a watchlist, replace its symbols, or delete a watchlist (cannot remove your last one) |
| Update Portfolio |
Deploy, undeploy, rename, delete, modify strategies, or change deployment frequency on an existing portfolio |
| List Datasets |
List your saved sandbox datasets with ids for follow-up delete or indicator wiring |
| Delete Dataset |
Soft-delete one owned sandbox dataset by id (does not delete promoted CustomIndicators) |
| List Custom Indicators |
List your owned CustomIndicators with ids for portfolio wiring or delete |
| Delete Custom Indicator |
Soft-archive one owned CustomIndicator by id (fork-safe; confirm when referenced by active portfolios) |
Subagents: Parallel Autonomous Exploration
For complex exploration tasks, Aurora can spawn subagents — independent AI agents that work in parallel. This is Aurora's most advanced capability.
How Subagents Work
When Aurora needs to explore multiple approaches simultaneously (e.g., testing different allocation splits or comparing different indicator parameters), she can:
- Spawn multiple subagents in parallel, each with a different task and optionally a different AI model.
- Wait for all subagents to complete their work (Aurora yields her worker slot while waiting, so system resources aren't wasted).
- Read the results from all subagents and synthesize a final answer.
Example: Strategy Exploration
If you ask Aurora to "Find the best rebalancing strategy for UPRO and GLD", she might:
- Spawn Subagent A: Test 30/70, 40/60, 50/50 splits with 30-day rebalancing
- Spawn Subagent B: Test the same splits with 60-day rebalancing
- Spawn Subagent C: Test the same splits with 90-day rebalancing
- Wait for all three to finish
- Compare all results and deliver the best-performing combination
Each subagent autonomously creates portfolios, runs backtests, and evaluates performance. The parent agent then synthesizes everything into a final recommendation.
Subagent Commands
| Command |
Description |
createSubagents |
Launch multiple subagents in parallel with different tasks |
waitForSubagents |
Wait for all child subagents to complete |
readAgents |
Get status and results from subagents |
stopAgents |
Stop specific subagents |
Choosing an AI Model
Aurora supports a variety of LLM models through OpenRouter and Requesty. You can select different models based on your needs and budget.
Model Tiers
| Tier |
Example Models |
Token Cost |
Best For |
| Budget |
Gemini 3.6 Flash, GPT-5.4 Mini, Kimi K2.6, MiniMax M2.5 |
1-3 tokens |
Quick tasks, screening, simple strategies |
| Advanced |
Gemini 3.1 Pro, GPT-5.4, DeepSeek V4 Pro |
25 tokens |
Complex reasoning, multi-step analysis |
| Premium |
Claude Opus 4.7, Sonar Pro Search |
40-50 tokens |
Deep research, strategy exploration, best quality |
Planning vs. Execution Models
During a multi-step agent run, Aurora uses two separate models:
- Planning model: Used to create the initial plan. A more capable model helps generate better plans.
- Execution model: Used during each step of the ReAct loop. Can be a faster/cheaper model for routine actions.
This dual-model approach lets you balance cost and quality — use a premium model for planning and a budget model for execution, or use premium for everything when quality matters most.
How to Choose a Model
You can configure which model Aurora uses from the chat or agent settings. Model costs are measured in research tokens per action, and pricing is dynamically updated.
Tip: For most users, the standard-tier models provide an excellent balance of quality and cost. Reserve premium models for complex agent tasks where reasoning quality matters most.
Research Tokens
Aurora uses research tokens to perform tasks. Each action (creating a strategy, running a backtest, researching a stock) consumes tokens. The number of tokens consumed depends on the model you select — budget models cost fewer tokens, premium models cost more.
| Plan |
Daily Research Tokens |
| Observer (Free) |
0 |
| Data-Driven Investor |
1,000 |
| Algorithmic Trader |
3,000 |
| Unstoppable Quant |
6,000 |
Tokens reset daily. Unused tokens do not roll over.
Tips for Getting Better Results
- Be specific: Instead of "make me money," say "Create a portfolio with 3 tech stocks that have strong fundamentals and use RSI for entry timing."
- Describe the whole goal: If your request involves multiple steps (research → create → backtest → optimize), just say so — Aurora automatically recognizes it needs multiple steps and handles it autonomously.
- Iterate: Start with a basic strategy, backtest it, then ask Aurora to refine it based on the results.
- Ask questions: Aurora can explain any concept, indicator, or result. Use her as a learning tool.
- Try different models: If you're not satisfied with the output quality, try a more capable model. If you want to conserve tokens, try a budget model for simpler tasks.
- Combine approaches: Use Aurora to generate initial ideas, then fine-tune them manually in the no-code UI.