> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hyreagent.fun/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat Agent

> How the Gemini-powered chat agent interprets questions and selects DeFi tools.

# Chat Agent

The Playground chat agent is powered by **Gemini 2.5 Flash-Lite** -- a fast, low-latency model optimized for tool selection and conversational DeFi analysis.

***

## Agent Capabilities

The agent can:

* **Auto-detect addresses** -- Paste a Solana token mint or wallet address and the agent identifies it and selects the right tool.
* **Multi-tool orchestration** -- Complex questions may trigger multiple endpoint calls in sequence.
* **Contextual follow-ups** -- The agent remembers conversation context for follow-up questions.
* **Explain results** -- Raw data is summarized in natural language with actionable takeaways.

***

## Tool Proposal Cards

Before executing a paid endpoint, the agent displays a **tool proposal card** showing:

* The endpoint name and description
* The price that will be deducted
* The parameters that will be sent
* An "Execute" button (or auto-execute if enabled)

This gives users visibility into what the agent is doing and how much it costs before any payment occurs.

***

## Model Details

| Property       | Value                                      |
| -------------- | ------------------------------------------ |
| Model          | Gemini 2.5 Flash-Lite                      |
| Provider       | Google AI                                  |
| Latency        | \~200-500ms first token                    |
| Context window | 1M tokens                                  |
| Streaming      | Yes (SSE)                                  |
| JSON mode      | Supported                                  |
| Cost           | $0.10 / $0.40 per 1M tokens (input/output) |

<Info>
  The chat agent model (Gemini) is separate from the tiered LLM cascade used to enrich API responses (Gemini, OpenServ SERV on decision endpoints, with OpenRouter/Venice fallback). The chat agent handles conversation flow and tool selection. The cascade handles data interpretation.
</Info>

***

## Example Conversations

<AccordionGroup>
  <Accordion title="Token research">
    **User:** What's the verdict on 7xKXtg2CW87d97TXJSDpbD5jBkheTqA83TZRuJosgAsU?

    **Agent:** I'll run a full token verdict analysis on this mint address.
    *\[Calls POST /trenches/token-verdict with mint parameter]*

    **Result:** This token has 847 holders with significant sniper concentration (top 5 wallets hold 34%). The bonding curve is at 72% completion. Dev wallet still holds 8%. **Signal: avoid** (confidence: 0.82).
  </Accordion>

  <Accordion title="Yield hunting">
    **User:** What are the best Meteora pools for SOL-USDC right now?

    **Agent:** I'll check pool recommendations for SOL-USDC.
    *\[Calls POST /lp/pools-recommend with base=SOL, quote=USDC]*

    **Result:** Top recommended pool is the SOL-USDC 1bp bin step pool with 127% APR and \$4.2M TVL. Suggested strategy: spot distribution, range +/-5%. **Signal: add\_liquidity** (confidence: 0.91).
  </Accordion>

  <Accordion title="Wallet intelligence">
    **User:** Is this wallet worth following? vhu9Nc7EQi6dpvk9R6GyrDjgrgnR2oyfRAeeYVmQCAp

    **Agent:** I'll pull a full wallet intelligence profile.
    *\[Calls POST /traders/wallet-intel with address parameter]*

    **Result:** 30-day profile shows a sniper archetype with 73% win rate on meme coins. Average hold time: 4.2 hours. Net PnL: +\$12,400. Copy-worthiness score: 7.8/10. **Signal: follow** (confidence: 0.88).
  </Accordion>
</AccordionGroup>

***

## Rate Limits

The chat agent has no artificial rate limits. Query frequency is bounded only by your escrow balance. The underlying Gemini model processes requests in real-time with sub-second latency.