[Enhancement]: Allow fully self-hosting the Code Interpreter (no API subscription required)

[CretegnyCelia]

What features would you like to see added?

• Ability to self-host the Code Interpreter without requiring a LibreChat subscription.
• Provide a local execution backend (e.g. Docker container, sandbox, or static-browser-server).
• Add configuration options to choose between local interpreter and subscription service.
• Ability to disable the subscription-based version if self-hosted backend is available.

More details

Summary

Currently, using LibreChat's Code Interpreter feature requires a paid API subscription. It would be great to support a fully self-hosted version of the Code Interpreter—just like all other parts of LibreChat—so users can run code locally without depending on the remote service.

Background

LibreChat is open-source and self-hostable, but the Code Interpreter requires a subscription and uses a closed-source API. This limits deployment flexibility, experimentation, and control.

Proposed Solution

• Provide an optional self-hosted code execution sandbox, similar to how Sandpack / static-browser-server works, or simple local containers.
• Provide documentation and configurations to set it up easily (e.g. Docker, domain, Caddy/Nginx, env vars).
• Allow disabling subscription-based execution via configuration, replacing it with a local interpreter.

Why this matters

• Security & compliance: Sensitive code and data never leave the local infrastructure.
• Privacy: Prevents exposure of internal or confidential information to third-party services.
• Control: Teams can fully audit and monitor the execution environment.
• Deployment flexibility: Works in air-gapped or restricted networks where external APIs aren’t allowed.

References:

• Related request to disable Code Interpreter in UI: [Enhancement]: Option to disable Code Interpreter in normal Chat #7238 :contentReference[oaicite:8]{index=8}
• Request to hide the button via config: [Enhancement]: hide the code interpreter button from chat bar #7165 :contentReference[oaicite:9]{index=9}

Which components are impacted by your request?

Endpoints

Pictures

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

I'm not against it being OSS one day, but for now, this is why it's not: https://www.librechat.ai/docs/features/code_interpreter#why-a-paid-api

[yoshirocks7]

Your reasoning makes ZERO sense since you've been charging us monthly without responding to ANY of our questions for support to even get Enterprise up and running. I get you want/need the support and we don't have any issues with paying for it, but there is an extreme lack of support, communication, execution on your team. I've written countless emails and scheduled appointments with you all to be either unanswered or cancelled at the last minute. If you don't want to support the project, then don't. Unlock Librechat so we can use other options. Without it, the usability of Librechat goes way down for more advanced users and is only as good as what the Models can provide.

[Nowaker]

I've absolutely no issue paying up...

[leondape]

Me neither!
It just needs to have all the pieces in place to be deployed in an enterprise environment. But as far as I understood they are working to re-launch the Coder Interpreter soon

[khyjb1995]

The subscription still unavailable till end of april

[CretegnyCelia]

Thank you for the quick response ! That makes sense and I appreciate you taking the time to clarify.

[Andrew-Chen-Wang]

I'm aiming to make one myself with Pulumi configuration. I tried messaging them for self-hosted, enterprise, willing to pay, but no one got back. Happy to share once I need beta testers, though it's likely going to be based on someone's existing solution like https://github.com/engineer-man/piston?tab=readme-ov-file or http://judge0.com/

[malikmalikayesha]

Hey @Andrew-Chen-Wang, i hope you are well!
My question is, why didn't you use Piston or Judge0 directly? Why do you want to build a system on top of them? Is there any setback using them directly?

[leondape]

I think Andrew's system on top "pulumi" ist just for infra as a code management. It would use pistons/judge0 directly.
I missed this recommendation. It would be cool to have it like this: Self deploy any Code interpreter you like but having the freedom in librechat to select wether to use:

• Librechat Code interpreter
• Judge0
• Piston
  always with options to use cloud or self hosted.

This would require the alternative code executors to be embedded into librechat just as deeply as librechat OG code interpreter with file support, UI support and response streaming.

I'm wondering if the librechat team would allow that PR because that would undermine the income stream via Librechat Code interpreter. On the otherhand the project and devs have been really embracing the Open Source mentality hard so they I believe they would be open for that.

May the best product win!

EDIT:

Piston could be integrated with full feature parity using its native files[] array for uploads and /api/v2/execute API.
Main challenge: Piston doesn't natively capture generated output files (Librechat Code Interpreter seems to do this very nicely).
Workaround: Return output files as base64-encode. - Add Decode step on Librechat side.
Open questions:
Session management (how to maintain state across conversation turns?)
Security: Didn't dive into librechat security implications yet.

Note: Judge0 has the same file capture limitation, so implementation effort is comparable for both.

[Andrew-Chen-Wang]

Hey, so the two links I recommended are simply environments to execute code in a sandbox safely.

Pulumi is to help users deploy those sandbox platforms into production and help the spin up and down.

As mentioned, LibreChat's source code expects a certain API schema and response structure that's incompatible with these other services, so you'll have to build your own API service to communicate with the sandbox and LibreChat. And as mentions as well, this is an income stream that may or may not be opened up.

Sadly, I've moved on from this. There is another open source chat solution that runs code in the browser but I think that is a bit limiting.

[leondape]

I think it's not so impossible.
We could re-use a lot of the UI and logic components just adapt for Piston.
The only thing, which is "ugly" is output file and download handling.

Regarding Librechat's openness to this, it would be great to hear the devs opinion about this before we start an implementation.
I think Librechat Codeinterpreter would be the superior product (nice file support) if we could deploy or purchase it nicely.

I could imagine there is openness, I would really dislike having a fork of LC just for this reason.
@danny-avila what do you think? Would you allow a PR to support other code sandboxes than Librechat Code Interpreter?

[ronniekolk]

Hi @danny-avila ,

we originally planned to use the paid version of CodeInterpreter.

Unfortunately, we haven't received any response, either through the contact form on the website or via the "Enterprise" button (which is no longer visible). Our private messages on Discord have also gone unanswered.

Our main challenge here in Germany is that we are only allowed to use solutions that comply with the GDPR (General Data Protection Regulation). If CodeInterpreter meets or can be made to meet these requirements, we would be happy to implement it—the need is definitely there.

Is there any current information available, or a way to contact you to discuss this further?

[Dannypeja]

+1!

I would be very happy to pay a very fair amount for the code interpreter but I cannot roll it out for my company as long as we don’t get it GDPR compliant.
I‘ve Repeatedly tried to reach out about that matter and enterprise implementation but unfortunately never received any response!

I fully understand @danny-avila and team are very much occupied by all other matters but it would be nice to know if any support in that direction is possible. The code interpreter pages suggests such solutions can be worked out but I really don’t have the feeling there is capacity to help me.

Could you clarify that?

Many thanks!

[yoshirocks7]

+2!

Also very frustrated with lack of support for enterprise code interpreter. Have been trying for months with little to no response on implementation. If a paid version can't be supported, Librechat should allow for integration of other interpreters since it's a feature that is a necessity if users want to do any type of data analysis.

[usnavy13]

https://github.com/usnavy13/LibreCodeInterpreter

[surak]

Hi Joseph, I am not sure how to point Librechat to it

[usnavy13]

set LIBRECHAT_CODE_BASEURL= in your env and set LIBRECHAT_CODE_API_KEY= to the one you created

[leondape]

Here are PRs for an alternative solution using Piston:

#11159
LibreChat-AI/librechat.ai#477

[leondape]

I think i solved this now more cleanly with Librechat Judge0 Bridge
This simple proxy server enables you to use Librechat Code Interpreter with Judge0 in under 30 seconds with no code changes, no merge conflicts.
As far as I see it there is 100% feature parity including file usage.
Please remark that, for now, this only works with Judge0 hosted, but probably soon will work for self-hosted as well.

[usnavy13]

https://github.com/usnavy13/LibreCodeInterpreter

[thehapyone]

Here is an OSS Code Interpreter anyone could use - https://github.com/thehapyone/code-interpreter, supporting a couple of languages

[danny-avila]

We will be open sourcing the code interpreter by the end of q1 this year: https://www.librechat.ai/blog/2026-02-18_2026_roadmap#dynamic-context

[Husamdien]

checking also @danny-avila

[CampbellReid]

Any update? Thank you

[leondape]

Skills are now developing in Frontend. For full support of PTC they will need to have the Code Sandbox ready. So I assume it won't take much much longer :)

[MaciejHajdenrajch]

Hello, are there any updates? Many thanks in advance!

[danny-avila]

Aiming to have progress on this by end of this week. Will update here as things land. Appreciate everyone's patience.

[Leon135]

[usnavy13]

https://github.com/usnavy13/LibreCodeInterpreter while you wait...

[sirsimonson]

I'm using your codeinterpreter which worked great. Until the most recent update where run_code_with_bash was introduced. Is there a known workaround?

@danny-avila Any updates on the timeline of the official version?

[Roboman124]

I'm using your codeinterpreter which worked great. Until the most recent update where run_code_with_bash was introduced. Is there a known workaround?

@danny-avila Any updates on the timeline of the official version?

Hey sorry to bother I was trying to setup it but was having issues with the api key is that the issue your mentioning the tool breaking it or am i dumb?

[Leon135]

@sirsimonson . The thing is that currently ut works only in bash tool... That is because librechat changed their execution steps. Python/js etc. work only in the interactive REPL mode. That's actually all right, most of the chatbots do the same especially claude, using heredoc to write python file, then run it.

[sirsimonson]

@Leon135 I tried by asking out my agent how execution could be possible and yes and got it working so far using bash tool. Anyways it's poor userexperience and im excited for the official release.

@Roboman124 No I worked around that by simply forcing it to not use JWTs with a little code adjustment. See here: #13114

[jmaddington]

Has the offical one still not been released?

[danny-avila]

You can find the project here once it goes public, waiting on security to flip the switch: https://github.com/ClickHouse/code-interpreter