llmsh

A terminal chat and repo summarization interface for LLMs using PIG inference gateway.

The interactive chat and ask flows enable direct interactions with the models servred through PIG. summarize analyzes and reduces large files or entire folders based on user instructions.

Install

Requires Python 3.12+ and uv.

git clone https://github.com/hosamsh/llmsh
cd llmsh
uv sync
uv tool install .

Usage

Run llmsh to open the interactive chat. First run walks you through setup.

• llmsh interactive chat tui for day-to-day work with PIG models.
• llmsh ask one-shot path for quick prompts in scripts or shell workflows.
• llmsh summarize for analyzing files, folders, and codebases with map-reduce style summarization.

CLI commands

llmsh                  Interactive chat (default)
llmsh ask <prompt>     One-shot prompt, prints response to stdout
llmsh summarize <path> <instruction>   Analyze files or folders with map-reduce summarization
llmsh doctor           Run diagnostics on current config
llmsh profile list|add|use|set-model|remove   Manage profiles
llmsh endpoint list|add|remove                Manage API endpoints
llmsh session list|show|delete                Manage saved sessions

ask options: --stdin, --json, --profile <name>, --system <text>

Slash commands (inside the chat)

/help                  Show command reference
/clear                 Clear the transcript
/retry                 Re-send the last message
/copy                  Copy last assistant message
/save                  Save current session
/load [id]             List or load a session
/session               Session management (list/save/load/delete)
/profile               Manage profiles (list/add/use/set-model/remove)
/endpoint              Manage endpoints (list/add/remove)
/doctor                Run diagnostics
/summarize <path> <instruction>   Analyze files or folders with map-reduce summarization
/cancel                Cancel active flow or request

Configuration

Config file: ~/.config/llmsh/config.toml

Repository example: config.example.toml

Set up interactively via llmsh (onboarding) or llmsh endpoint add / llmsh profile add from the command line. Or edit the TOML directly:

current_profile = "default"

[endpoints.local]
base_url = "http://localhost:8006/v1"
auth_mode = "none"

[profiles.default]
endpoint = "local"
model = "your-model-name"
max_tokens = 1024

For endpoints requiring API keys:

[endpoints.openai]
base_url = "https://api.openai.com/v1"
auth_mode = "api_key"
api_key_env = "OPENAI_API_KEY"

Development

uv sync --extra dev
uv run pytest
uv run ruff check src/ tests/
uv run ty check src/

License

MIT — see LICENSE.