spark 1.5 . by forgewright-ai . github.com/forgewright-ai/spark

spark -- credits and licenses

spark vendors none of the projects below. bootstrap.sh downloads each one, pinned by version and sha256, from its own upstream to your machine at install time; apt and Homebrew install the rest from their own repositories. spark's own code is MIT (LICENSE). The banner in home/.config/spark/banner is spark's own artwork.

The engine

llama.cpp -- https://github.com/ggml-org/llama.cpp -- MIT (c) The ggml authors

Release b10689 (LLAMA_VERSION in bootstrap.sh), six flavours (macOS arm64, macOS x64, Linux x64, Linux x64 Vulkan, Linux arm64, Linux arm64 Vulkan), each pinned by sha256.

The prompt

starship -- https://github.com/starship/starship -- ISC (c) Starship Contributors

Version 1.26.0 (STARSHIP_VERSION in bootstrap.sh), pinned by sha256 on Linux; Homebrew installs the same tool on macOS with no version pin.

micro's aspell plugin, installed by micro -plugin install aspell from micro's own plugin channel -- not pinned by spark; the shell layer only (spark shell on).

Fonts

JetBrains Mono -- https://github.com/JetBrains/JetBrainsMono -- SIL Open Font License 1.1, Reserved Font Name "JetBrains Mono"

Nerd Fonts -- https://github.com/ryanoasis/nerd-fonts -- MIT (the patcher)

Nerd Fonts patches JetBrains Mono and renames the result "JetBrainsMono Nerd Font" -- version 3.5.1 (NERDFONT_VERSION in bootstrap.sh), pinned by sha256 on Linux; Homebrew's font-jetbrains-mono-nerd-font cask installs the same family on macOS with no version pin.

Palettes

Colour values only, no code copied; MIT, each:

Shell tools (apt / Homebrew, spark shell on)

None of these is shipped by spark; apt and Homebrew install them from their own repositories, unpinned.

Models

Downloaded by bootstrap.sh, never vendored. Each row's size and sha256 come from Hugging Face's file metadata.

The curated list (models.env) holds open licenses only: Qwen3 1.7B, 4B, 8B, 14B and 30B-A3B, by Alibaba's Qwen team, Apache-2.0. GGUF quantizations by unsloth (1.7B) and bartowski (the rest).

The embers list (embers.env): Qwen2.5-Coder-7B-Instruct and Qwen3-4B-Thinking-2507, Apache-2.0, quantized by bartowski.

The community list (community.env) may name any license, and names it per row: gemma-3-12b-it, by Google, is under the Gemma Terms of Use (https://ai.google.dev/gemma/terms), not an open-source license; spark asks before downloading a community row.

Built with Claude

The v1.0 refactor and simplification -- the layer split, the guided first run, the model catalog, the chooser, the chat, the fresh-account proofs, and these docs -- were designed and implemented with Claude (Anthropic) in Claude Code, directed and reviewed by the maintainer.

Every such commit carries a Co-Authored-By: Claude trailer, so git log tells the same story as this paragraph. The mistakes are the maintainer's.

Corrections

A correction or a missing name is a pull request away.