Skip to content

Releases: Tok/impulse-instruct

v0.7.4

13 Apr 16:36
@github-actions github-actions
v0.7.4
a56a915

Choose a tag to compare

View all tags
v0.7.4 Latest
Latest

Impulse Instruct v0.7.4

A smart synthesizer with a virtual production team inside — multiple LLM agents collaborate to write patterns, shape sound, and evolve tracks in real time.

Quick start

  1. Download the impulse-instruct-v0.7.4-<platform>.zip for your platform (not the auto-attached "Source code" zip).
  2. Extract — you'll get the binary, configs, start scripts, and an empty models/ folder.
  3. Get a model. Either download gemma-4-E4B-it-Q4_K_M.gguf manually from unsloth/gemma-4-E4B-it-GGUF and drop it in models/, or run ./download-models.sh / download-models.bat.
  4. Run ./start.sh (Linux) or double-click start.bat (Windows) — this launches the app in a console so logs stay visible.

Windows note: the .exe is not code-signed. SmartScreen may show "Windows protected your PC" — click More info → Run anyway. The build is reproducible and attested; see verification below.

Verify

sha256sum impulse-instruct-linux-x86_64
gh attestation verify impulse-instruct-linux-x86_64 --repo Tok/impulse-instruct

System requirements

  • Linux: libasound2, libxcb (standard on most distros)
  • Windows: Windows 10+ x64, MSVC runtime (usually pre-installed)
  • LLM inference: NVIDIA GPU recommended; CPU inference works but is slow

Recommended models

  • Gemma 4 E4B Q4_K_M (~4.6 GB) — best accuracy, default; needed for all presets
  • Bonsai 8B Q1_0_g128 (~1.1 GB) — lightweight agent for multi-model teams

Full Changelog: v0.7.3...v0.7.4

Assets 6
Loading

v0.7.3

12 Apr 20:16
@github-actions github-actions
v0.7.3
6ea5397

Choose a tag to compare

View all tags
v0.7.3

Impulse Instruct v0.7.3

A smart synthesizer with a virtual production team inside — multiple LLM agents collaborate to write patterns, shape sound, and evolve tracks in real time.

Quick start

  1. Download the .zip for your platform
  2. Extract — you'll get the binary, configs, start scripts, and a models/ folder
  3. Run scripts/download-models.sh (or .bat) to fetch Gemma 4 E4B (~4.6 GB)
  4. Run ./start.sh (Linux) or start.bat (Windows) — this launches the app with terminal logging

Verify

sha256sum impulse-instruct-linux-x86_64
gh attestation verify impulse-instruct-linux-x86_64 --repo Tok/impulse-instruct

System requirements

  • Linux: libasound2, libxcb (standard on most distros)
  • Windows: Windows 10+ x64, MSVC runtime (usually pre-installed)
  • LLM inference: NVIDIA GPU recommended; CPU inference works but is slow

Recommended models

  • Gemma 4 E4B Q4_K_M (~4.6 GB) — best accuracy, default; needed for all presets
  • Bonsai 8B Q1_0_g128 (~1.1 GB) — lightweight agent for multi-model teams

Full Changelog: v0.7.2...v0.7.3

Loading

v0.7.2

12 Apr 14:27
@github-actions github-actions
v0.7.2
ce7189e

Choose a tag to compare

View all tags
v0.7.2

Impulse Instruct v0.7.2

A smart synthesizer with a virtual production team inside — multiple LLM agents collaborate to write patterns, shape sound, and evolve tracks in real time.

Quick start

  1. Download the .zip for your platform
  2. Extract — you'll get the binary, configs, start scripts, and a models/ folder
  3. Run scripts/download-models.sh (or .bat) to fetch Gemma 4 E4B (~4.6 GB)
  4. Run ./start.sh (Linux) or start.bat (Windows) — this launches the app with terminal logging

Verify

sha256sum impulse-instruct-linux-x86_64
gh attestation verify impulse-instruct-linux-x86_64 --repo Tok/impulse-instruct

System requirements

  • Linux: libasound2, libxcb (standard on most distros)
  • Windows: Windows 10+ x64, MSVC runtime (usually pre-installed)
  • LLM inference: NVIDIA GPU recommended; CPU inference works but is slow

Recommended models

  • Gemma 4 E4B Q4_K_M (~4.6 GB) — best accuracy, default; needed for all presets
  • Bonsai 8B Q1_0_g128 (~1.1 GB) — lightweight agent for multi-model teams

Full Changelog: v0.6.4...v0.7.2

Loading