Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.dzap.io/llms.txt

Use this file to discover all available pages before exploring further.

A small vocabulary covers most of DZap AI. Read this once; everything else fits into the same five buckets.

​
Agent

The Agent is the LLM-powered reasoning loop. Given a user query and context, it picks tools, calls them in sequence, and returns a final answer. DZap ships one production agent — ZapBot (CLIAgent in code) — backed by OpenAI by default. You can swap providers via env. See ZapBot.

​
Tool

A Tool is a typed function the agent can call. ~20 tools ship in the box: balance, price, swap-link, perform-zap, web-search, RAG-over-docs, scheduler, etc. Each is a tiny adapter between the agent and a backend API or external service. See Tools. Anatomy of a tool: 
{
  name: 'PriceTool',
  description: 'Fetch USD price for tokens',
  parameters: { tokenAddresses: 'string', chainId: 'number' },
  execute: async (args) => { ... }
}

​
Skill

A Skill is a pre-packaged context bundle for an LLM editor (Cursor, Claude Code). It primes the model with DZap’s API conventions so completions match real endpoints. Three published skills today: dzap-sdk, dzap-fuse-api, dzap-trade-api. See Skills. Skills ≠ tools:
SkillTool
Static markdown contextLive function the agent can call
Loaded into editor at session startInvoked by the agent at runtime
For human-driven codingFor agent-driven actions

​
Memory

Memory persists conversation state across turns within a session, plus optional long-term memory across sessions. Two layers in code:
  • ChatHistory — short-term, in-memory message log
  • Memory — long-term store for facts the agent should recall
For most apps, default memory is fine. Customize when you embed the agent in a multi-tenant product.

​
Scheduler

A persistent task scheduler (SheluderTool, GetAllSchedulesTool) backed by SQLite. Lets the agent schedule actions (“DCA $100 of ETH every Friday at 9am”). Underlying record format is documented in the tools reference.
The scheduler is implementation-complete but interface-evolving. Use cases stabilize through 2026.

​
Putting it together

User → Agent → Tools → DZap APIs / external services
              ↑
              └── Memory (per-session + long-term)
              ↑
              └── Skills (loaded into editor, separate from runtime)

​
Next

ZapBot deep dive

Reasoning → tool calls → intent → execute.

Vibe coding

Prompt patterns and gotchas.