Little-Coder
A terminal agent for smaller models
Little-Coder is a coding agent built on pi, with extensions aimed at working with smaller local language models. It runs in a project directory and provides tools for reading files, making edits and running shell commands. Its documented model configuration covers llama.cpp, Ollama and LM Studio, with cloud providers also available.
This is a developer tool that needs a working model connection. The README specifies Node 22.19.0 or newer; installing the agent does not start a local inference server. Choose a model and make sure its server is reachable before expecting the terminal session to answer requests. Hardware requirements depend on the model and server configuration.
Research a change before implementing it
Plan Mode separates research and planning from implementation. It can use read-only sub-coders to investigate a task, ask clarifying questions and present a plan. Approval saves the plan, while /implement starts the implementation phase in a fresh session.
Planning and implementation can use different configured models. Leaving the plan and action model settings unset keeps both phases on the active model. On a single local backend, switching models can involve unloading and loading weights, so choose that split with your setup in mind.
Understand the permission boundary
The default permission mode checks shell commands against allowed prefixes and additional rules for chained commands, redirects and inline interpreter code. Manual mode asks before shell commands and file writes or edits.
The shell whitelist is a speed bump, not a sandbox. The README explicitly recommends a container, VM or human approval when a stronger boundary is needed. Review edits and commands with the same care you would give any coding agent.
Add extensions deliberately
The launcher loads its bundled extensions and provides opt-in paths for adding your own. The extension guide describes a dedicated configuration directory and the /extensions view for checking what loaded.
Extensions placed in the dedicated directory load after the bundled set. Enabling pi's broader extension discovery can also allow project-local extensions, so inspect that configuration before using an unfamiliar repository.