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.

ZapBot is the agent that interprets your prompt, picks tools, and executes intents. It powers the CLI, the SDK-AI runtime, and the MCP server.

What it is

In code: CLIAgent in DZapAI. A tool-calling LLM agent built on the Vercel AI SDK with auto-retry and timeout handling. In production: the brain behind:
  • The dzapai CLI shell
  • DZapSDK.ask() calls
  • MCP-exposed actions in Claude Desktop / Cursor

What it can do

CategoryExamples
ReadGet balance, price, pool data, news, sentiment
PlanBuild a swap/zap path with fees and slippage
ExecuteSign + send via session-scoped key (with confirmation)
SchedulePersist recurring actions via the scheduler

Default configuration

SettingDefaultOverride
Modelgpt-4o-miniDZAP_OPENAI_MODEL=gpt-4o
Max retries2DZAP_MODEL_MAX_RETRIES
Timeout30sDZAP_MODEL_TIMEOUT_MS
ProviderOpenAI(other providers via Vercel AI SDK)

Three things to know

  1. Confirmations matter. For execution paths (swap, zap), ZapBot surfaces a plan and waits for confirmation. Configure in Safety.
  2. Session-scoped keys. Never load a wallet key into long-lived process memory. ZapBot uses session keys that die with the process.
  3. Tool calls are auditable. Every call is logged with inputs and outputs in result.toolCalls.

Next

How it works

Reasoning → tools → intent → execute.

Safety

Session keys, confirmations, limits.