Support Codex on ubuntu24.04 on RISC-V (riscv64)
What version of Codex is running?
codex --version file:///home/joakim/.npm-global/lib/node_modules/@openai/codex/bin/codex.js:59 throw new Error(Unsupported platform: ${platform} (${arch})); ^ Error: Unsupported platform: linux (riscv64) at file:///home/joakim/.npm-global/lib/node_modules/@openai/codex/bin/codex.js:59:9 at ModuleJob.run (node:internal/modules/esm/module_job:263:25) at async ModuleLoader.import (node:internal/modules/esm/loader:540:24) at async asyncRunEntryPointWithESMLoader (node:internal/modules/run_main:117:5) Node.js v20.19.0
What subscription do you have?
ChatGPT plus
Which model were you using?
none - I do not even get there.
What platform is your computer?
Linux 6.6.92-eic7x-2025.07 riscv64 riscv64
What issue are you seeing?
I cannot use Codex on my ubuntu24.04/riscv64 host. I do get this error:codex --versionUnsupported platform: ${platform} (${arch})`);
file:///home/joakim/.npm-global/lib/node_modules/@openai/codex/bin/codex.js:59
throw new Error(
^
Error: Unsupported platform: linux (riscv64)
at file:///home/joakim/.npm-global/lib/node_modules/@openai/codex/bin/codex.js:59:9
at ModuleJob.run (node:internal/modules/esm/module_job:263:25)
at async ModuleLoader.import (node:internal/modules/esm/loader:540:24)
at async asyncRunEntryPointWithESMLoader (node:internal/modules/run_main:117:5)
Node.js v20.19.0
`
What steps can reproduce the bug?
Run Codex on ubuntu24.04 / riscv64.
What is the expected behavior?
_No response_
Additional information
_No response_
9 Comments
We don't provide a binary for riscv64 currently. I've changed this from a bug report to an enhancement request.
Ok! I managed to build one - works fine - so should not be a problem.
@joakimeriksson upstream it?
@etraut-openai why not you just have to order a few machine of #RISCV and take @joakimeriksson upstream patches, there will be 10K people using very powerful RISCV laptops with ubuntu 26.04 next year in 2026 as Deepcomputing 100 universities/AI startups/ Community Projects support programmes announded in RISCV Summit NA
https://deepcomputing.io/deepcomputing-at-risc-v-summit-na-2025-empowering-the-future-of-open-ai-computing/,
hope you wont get 10K issues raised and swarmed on github.
https://store.deepcomputing.io/products/dc-roma-ai-pc-risc-v-mainboard-ii-for-framework-laptop-13
Many thanks in advance.
Thanks for the suggestion. Please upvote this issue if you want to see this. We use upvotes / reactions to inform priorities for enhancement requests.
@etraut-openai how do we do upvotes please? @joakimeriksson do you know how? Codex definitely can be one of our AI open source project in our #RISCV hardware support program
So these were the few changes that I needed to do to get it building on riscv64.
https://github.com/joakimeriksson/codex/tree/riscv64-support
(in codex-rs - do cargo build).
Click on the reactions button (the little smiley face beneath the post) and add a "thumbs up" reaction.
With the help of Codex, I completed a functional reference implementation for Linux/riscv64 support:
https://github.com/sudaoer/codex/tree/riscv64gc-review
The current branch is intentionally limited to the
riscv64gc-unknown-linux-muslrelease path and is split into three commits for easier review:rgrusty_v8release artifactsOne platform-specific issue I found is V8 sandbox reservation on Sv39 systems. V8 normally attempts to reserve a 128 GiB address-space region; under Sv39 this cannot be fully reserved, so V8 falls back to a partially reserved sandbox. Code mode still works, but the address-space isolation is weaker than on systems with a sufficiently large virtual address space. I've documented this as a known platform limitation.
I have access to real RISC-V Linux hardware and can test runtime behavior and any changes needed during review.
If upstream RISC-V support is a direction the project is interested in, I'm happy to adjust the implementation or split it further according to maintainer preference.