● DOC.AGENT.023AGENTS / TOKEN ANALYST
[ TOKEN ANALYST ]
AI with context
Upload your knowledge base. The agent retrieves the relevant chunks, combines them with live on-chain data, and writes a short verdict: bullish, neutral, or bearish with reasons.
[ 01 ]
How a signal is produced
- 01KNOWLEDGE
Upload what the AI should know
Paste text, docs, tweets, or URLs. Everything gets chunked and embedded via OpenAI text-embedding-3-small (1536-dim) into the pgvector store
- 02TRIGGER
Set a trigger
New token crossing liquidity, watchlist entry, or tag match
- 03RETRIEVE
Retrieval into prompt
On fire, the agent pulls the most relevant knowledge chunks and injects them into the LLM prompt alongside live on-chain stats
- 04VERDICT
Written verdict
Output: bullish, neutral, or bearish, plus three to five bullet reasons and a confidence number
※ Note
RAG (retrieval-augmented generation) is currently wired for Token Analyst agents. Other agent types run without custom knowledge injection for now.