BACKPACKalpha

Public alpha · Apache-2.0

Open models.
One runtime.

One open-source CLI that finds, verifies, and runs the right engine for each model.

View source ↗

Validated: Windows x64 Linux x64 preview macOS arm64 build preview

Backpack walking gorilla mascot
30BA REAL RUNLOCAL
$ backpack pull qwen3-coder-30b-a3b$ backpack run qwen3-coder-30b-a3b

ready · llama.cpp · local

What / Where / How / Who

One model.
Wherever you compute.

Choose the model and the compute. Backpack prepares the runtime and gives your tools one consistent way in.

WHAT RUNSOPEN MODELyou choose it
BACKPACKmodel + target = runtime
WHERE IT RUNS
LOCALVALIDATEDSSH / REMOTEEXPERIMENTALBACKPACK CLOUDCOMING SOON
WHAT USES IT
CLIAPICODEXCLAUDE CODE

Model is what. Target is where.The model stays independent from the hardware that runs it.

Backpack handles how.It resolves the runtime machinery; launch connects the tools that consume it.

Experimental tool bridge

Open models.
Inside your coding tools.

Backpack starts the tool you already use and points only that process at its local API.

BACKPACK LOCAL API
01EXPERIMENTAL
Codex
$ backpack launch codex
02EXPERIMENTAL
Claude Code
$ backpack launch claude
03EXPERIMENTAL
OpenCode
$ backpack launch opencode

Bring the agent yourself. Codex protocol/tool-loop coverage is ahead; Claude Code and OpenCode still require wider real-binary qualification.

The commands worth remembering

Pull.
Run.
Launch.

One vocabulary from model download to the tools that use it.

Command reference ↗
01 / GET THE MODELbackpack pull
02 / START THE RUNTIMEbackpack run
03 / OPEN YOUR TOOLbackpack launch

doctormodelspsstoptranscribespeak

A catalog built to grow

Open models.
Packed to run.

Text, code, transcription, and speech today. New modalities join only after their execution path is qualified.

Explore the model catalog →
01
TEXT + CODEQwen3-Coder / Qwen2.5 / SmolLM2
RUNNING
02
HEARWhisper / Qwen3-ASR
RUNNING
03
SPEAKKokoro
RUNNING
04
IMAGE + VIDEOZ-Image / Wan / more to qualify
NEXT

Open source / Cloud

Your compute.
Or ours.
Same Backpack.

One runtime architecture. The difference is who provides the compute.

01OPEN SOURCEAPACHE-2.0

Open models.
Your compute.

Run Backpack on hardware you control. No Backpack account required.

LOCALSSH / REMOTE / EXPERIMENTALOPEN MODELSCODEX / CLAUDE / EXPERIMENTALOPENAI-COMPATIBLE API
Get Backpack ↓GitHub ↗
02BACKPACK CLOUDCOMING SOON

Open models.
Our compute.

Managed Backpack inference without managing the underlying GPU or runtime yourself. Planned for Backpack Pro.

SERVERLESS INFERENCE / PLANNEDCLOUD MODELS / PLANNEDDEDICATED COMPUTE / FUTURETEAMS / FUTURE
Request Cloud access →PRIVATE PREVIEW INTEREST

Early alpha

Pull it.
Run it.

The installer selects the newest published release, warns for prereleases, verifies the archive checksum, and installs without elevation.

Inspect-first and pinned installs ↗

Then

$ backpack doctor
$ backpack models
$ backpack pull smollm2-135m
$ backpack run smollm2-135m
Execution-validatedInterfaces and behavior may change before beta.