Install

Pre-release

pocket llm is in active development. M0 (scaffold) and M2 (init + doctor) have landed. The other commands — wiki generate, build, chat, serve, import, eval — are stubs that return not implemented. See where we stand for the milestone-by-milestone state.

There is no released binary yet. The install path below assumes you're building from source.

Prerequisites

• Go 1.25+ — required by apps/cli (bumped for bubbletea v1.3.10 transitive deps).
• CMake + a C toolchain — required by apps/llama to build libllama.a. Skip-able for now via the pocketllm_llama_stub build tag while M1 is pending.
• task (go-task) — orchestrator.
• Optional: bun — only for the embedded dashboard build.

Platform	C toolchain
linux	apt install build-essential cmake
darwin	Xcode Command Line Tools + brew install cmake
windows	Visual Studio Build Tools (MSVC) or MSYS2 + MinGW-w64

Build from source

git clone https://github.com/deemwar-products/pocket-llm.git
cd pocket-llm

# CLI only (skips cgo / libllama since M1 is pending)
cd apps/cli
go build -o /usr/local/bin/local-agents ./cmd/local-agents

# Or with the function-layer stub (no C toolchain needed)
go build -tags pocketllm_llama_stub -o /usr/local/bin/local-agents ./cmd/local-agents

Once M1 lands, task build from the repo root will produce the same binary with llama.cpp linked statically.

Verify

local-agents --help        # lists 8 subcommands
local-agents doctor --json # 8 health checks; warns on missing model

doctor will report error: embedding missing and similar until you run local-agents init — that's expected on a fresh install.

Next

• Quickstart — scaffold your first project.
• Project folder layout — what init creates.
• Command reference — every flag.