REPL Interface

The Zana REPL is the primary way to interact with the coding copilot. It provides an interactive terminal session where you can ask the LLM to read, write, edit, and search your codebase.

Starting the REPL

With Ollama (Local Models)

using Zana

zana(ollama=true, model="qwen3")

With Anthropic Claude

using Zana

zana(claude=true)  # Reads ANTHROPIC_API_KEY from environment

With OpenAI

using Zana

zana()  # Reads OPENAI_API_KEY from environment

With Custom API

using Zana

zana(api_key="your-key", base_url="https://api.example.com/v1", model="your-model")

Full Parameter List

zana(;
    api_key = nothing,             # API key (or from ENV)
    model = nothing,               # Model name (default depends on backend)
    base_url = "https://api.openai.com/v1",  # API base URL (OpenAI backend)
    ollama = false,                # Use Ollama backend
    claude = false,                # Use Anthropic Claude backend
    host = "http://localhost:11434",  # Ollama host
    use_openai_compat = false,     # Use Ollama's OpenAI-compatible endpoint
    working_dir = pwd()            # Working directory for tools
)

REPL Commands

All commands start with /:

/clear

Resets the conversation history and turn count. The LLM will have no memory of previous exchanges. Useful when switching topics or starting a new task.

/history

Displays a condensed summary of the conversation:

• User messages — Shows first 80 characters of each message
• Assistant messages — Shows tool calls or text preview
• Tool results — Omitted for brevity

/tokens

Shows cumulative token usage across all turns:

  Total tokens: 15420 in / 3280 out
  Turns: 5

/cd PATH

Changes the working directory for all tools. The system prompt and tool paths are updated automatically:

zana> /cd /path/to/other/project
  Working directory: /path/to/other/project

Multi-Line Input

Use a trailing \ to continue input on the next line:

zana> Write a function that \
  ...> takes a filename and \
  ...> returns the line count.

The continuation prompt ...> appears automatically. Lines are joined with newlines before being sent to the LLM.

How It Works

Each time you send a message, Zana follows this process:

1. User message is added to the conversation history
2. Full context (system prompt + conversation history) is sent to the LLM
3. The LLM may request tool calls (read a file, run a command, etc.)
4. Zana executes each tool and feeds results back to the LLM
5. Steps 3-4 repeat until the LLM produces a final text response (up to 50 rounds)
6. The response is displayed along with a summary of tool executions

Tool Execution Summary

When tools are used, Zana displays a summary:

  ⚙ glob_files  **/*.jl  (12ms)
  ⚙ read_file  src/Zana.jl  (3ms)
  ⚙ read_file  src/agent.jl  (2ms)

This project is a Julia package called Zana.jl...

  [tokens: 2450 in / 180 out]

Tips