Local model plus local tools
OpenJet is not just chat. It ships file tools, edit tools, search, directory listing, shell execution, and approval handling in the same terminal workflow.
Comparison
Local Alternative to Claude Code
If you want the terminal-native agent workflow, but you want the model path to stay local and have the stack set up for you instead of using a hosted coding service, this is the gap OpenJet is aimed at.
Local setup handled for you • Explicit approvals • Resume checkpoints • Benchmark sweeps
The actual difference
OpenJet is built around getting the local setup working for you.
It keeps the agent workflow on your machine instead of routing it through a hosted coding service.
It is designed for constrained hardware, edge devices, and sessions you need to resume later.
It is a coding agent with approvals and local tools, not just a local model wrapper.
Comparison
Where OpenJet differs in practice
This comparison uses the OpenJet source/docs in the adjacent repo and Anthropic's official Claude Code docs for the Claude side.
| Category | OpenJet | Claude Code |
|---|---|---|
| Model runtime | Sets up a local coding stack for you and keeps the agent path on your machine. | Official Claude Code docs position it around Anthropic-hosted usage, with AWS Bedrock and Google Vertex AI as enterprise options. |
| Install path | Clone the repo, run ./install.sh, then use open-jet --setup to reuse what is already there or provision the missing pieces for you. | Official install flow uses npm or the native installer, then account login and internet-backed AI processing. |
| Command execution | Shell commands require explicit approval before execution. | Claude Code also uses a permission model with read-only defaults and approval for bash commands. |
| Session recovery | Built-in /resume restores saved chat state and prefers resume checkpoints so you can continue later. | Claude Code documents conversation compaction and resume support, but OpenJet exposes explicit local session state and checkpoint files under .openjet/state/. |
| Constrained hardware | Setup is designed for real local hardware and the product includes benchmarking so you can tune performance on the same machine you use it on. | Official docs describe standard developer-machine setup, not local model fitting or hardware-specific benchmarking. |
| Built-in device access | Includes device_list, camera_snapshot, microphone_record, and gpio_read plus @camera0/@mic0/@gpio0 tagging in chat. | Claude Code documentation emphasizes integrations and permission controls. OpenJet ships built-in local camera, microphone, and GPIO-aware tools directly in the product. |
Why OpenJet
Why pick OpenJet instead of just using a hosted coding agent?
Setup handles the ugly parts
The setup flow is there so you do not have to manually assemble the local stack before you can start using the agent.
Actually built for interrupted sessions
OpenJet persists chat/session state, stores resumable checkpoints, and documents where manifests, transcripts, and KV cache snapshots live on disk.
More than coding
For edge and embedded work it can tag and use cameras, microphones, and GPIO-backed sources, which is a different shape of product than a cloud coding agent.
Decision
Pick OpenJet if the whole point is to stay local and have the setup handled for you.
If you want the hosted path, use a hosted product. If you want the coding-agent workflow on your own hardware without building the local stack by hand first, that is the job OpenJet is trying to do.
Next step
See the offline AI coding agent page for the product positioning, or go to the docs if you want installation details.