leleka

Lightweight local-first AI assistant CLI — powered by Ollama.

Leleka brings the power of large language models to your terminal with zero cloud dependency. Run queries, hold conversations, and monitor your system health — all from a single command-line tool backed by your local Ollama instance.

The Stork is watching...

Features

• One-shot prompts — fire off a question with optional file context and get live-streamed responses
• Persistent chat sessions — multi-turn conversations saved as markdown files for later reference
• System monitoring — list local Ollama models, inspect system pulse events
• Project scaffolding — structured Markdown documents with templated chapters
• Workspace helpers — scrape context from CSV, JSON, and code files into LLM-ready format

Installation

From source

git clone https://github.com/<user>/leleka.git
cd leleka
uv pip install -e .

Dependencies

Package	Version	Purpose
typer	>= 0.12	CLI framework
rich	>= 13	Live streaming panels & tables
ollama	>= 0.3	Local LLM backend

Ollama must be installed and running locally with at least one model pulled (e.g. mistral:7b).

Usage

# Unified command — interactive chat (default)
leleka run

# One-shot prompt from piped text
cat file.txt | leleka run

# With model selection
leleka run --model mistral:7b

# System monitoring
leleka ps --models          # list local Ollama models
leleka ps --pulse           # display system pulse data

Environment variables

Variable	Purpose	Default
CONTEXT_PATH	Directory for context files	project root /context
CHATS_PATH	Directory for saved chat sessions	project root /chats
MODELS_PATH	Directory for model system prompts	project root /models

Architecture

leleka/
├── main.py              # CLI entry point — wires sub-apps together
├── config.py            # Path constants, defaults, logo text
├── core/
│   ├── llm.py           # Streaming engine (ollama.chat + Rich Panel)
│   └── projects_engine.py  # Markdown section dataclass (ProjectFile)
├── commands/
│   ├── run.py           # Unified prompt / chat command (new default)
│   ├── ps.py            # System monitoring command
│   └── ask.py           # Deprecated alias for `run`
└── tools/
    ├── session_ops.py       # Shared streaming/context helpers
    ├── projects_helpers.py  # Project scaffolding utilities
    ├── ps_helpers.py        # Ollama model listing, pulse display

Roadmap

• Unify ask + chat into a single command with stdin piping support
• Provider abstraction layer (swap Ollama for other backends)
• Type hints, docstrings, and test coverage
• Restore missing .markdown_utils submodule

License

[To be determined]