2025 Roadmap

[danny-avila]

Please share your feedback and thoughts on the 2025 roadmap for LibreChat 🎉

https://www.librechat.ai/blog/2025-02-20_2025_roadmap

[aleibovici]

Hey @danny-avila where is the 2025 roadmap? On the website I can only find the 2024.

[plasticlobster]

https://www.librechat.ai/blog/2025-02-20_2025_roadmap

[danny-avila]

I opened this thread before the blog went live 😅

[marlonka]

Roadmap makes me really excited for 2025, especially the gemini improvements and admin Features will be huge for us. One idea I still have would be some image playground within Librechat that can use Black Forest Labs API to generate and modify like on replicate playground or midjourney edit function.

[ColbySerpa]

MCP Marketplace / Store as seen in Cline & 5ire

There are a few other apps that have this MCP store, like 5ire and most recently Cline. It would make Librechat the leader, given the fork system and everything else in Librechat is so well thought-out.

[dvejsada]

@danny-avila looks great, we are looking forward to all the new features. As our company deployment seems to be quite successful, we will be happy to give LibreChat some publicity sharing our use case;)...

[hamletfg]

As a WebDev and QA Engineer student I've been using Librechat nearly every day. It's been absolutely wonderful having several AI models in one place so a huge thanks to you, Danny and the team for building a great product. I'll be happy to contribute to LibreChat in the future as my skills grow 🙌

[samuelpath]

@hamletfg I don't think you need to wait for your skill to grow before contributing. You can grow your skills by contributing :).

[hamletfg]

@samuelpath Thanks for the encouragement :) I'll check out the page on how to contribute, let me know if there are any other recommendations on how to start without being overwhelmed.

[hibukipanim]

Are there plans for supporting OpenAI's new Responses API which will also replace the Assistants API.

From https://openai.com/index/new-tools-for-building-agents/:

Based on developer feedback from the Assistants API beta, we’ve incorporated key improvements into the Responses API, making it more flexible, faster, and easier to use. We’re working to achieve full feature parity between the Assistants and the Responses API, including support for Assistant-like and Thread-like objects, and the Code Interpreter tool. Once this is complete, we plan to formally announce the deprecation of the Assistants API with a target sunset date in mid-2026. Upon deprecation, we will provide a clear migration guide from the Assistants API to the Responses API that allows developers to preserve all their data and migrate their applications. Until we formally announce the deprecation, we will continue delivering new models to the Assistants API. The Responses API represents the future direction for building agents on OpenAI.

Thanks

[danny-avila]

Are there plans for supporting OpenAI's new Responses API

Nope, we have our own agent framework: https://www.librechat.ai/docs/features/agents

[danny-avila]

And Assistants API will officially be deprecated following this announcement

[hibukipanim]

This is disappointing for our use cases, which we started to implement with LibreChat's Assistant API support (while Responses API would make it much easier and nicer).
TL;DR: In our use-case we are interested in implementing the server side of Responses API, and use LibreChat as the conversational UI against it.

The use case is that we develop various services, which internally perform complex LLM-augmented flows using various code frameworks like LangChain, LlamaIndex. These services can expose a conversational API so that users can interact with them from LibreChat. Unlike in the agent framework, the user has no control over which LLM is used (as a combination of them can be used behind the scenes) and which tools are enabled (as the services have this logic internally), and the entry point that handles each new user message is custom code in the service, not directly a specific LLM.
In order for users to be able to converse with such an assistant service using LibreChat, the service needs to implement an API which LibreChat works with. The Completions API is not good enough as it doesn't support streaming events about long-running backend execution, or return files and images. I wrote about it also here #5047 (comment). The Assistants API, while supporting our use-case, is too complex and stateful. The new Responses API seems as a much better fit for an API we'd implement the server side of.

[danny-avila]

@hibukipanim what you’re specifically asking for is outside of the scope of LC. Responses API is just another layer on top of completions so it can actually be used with LC Agents framework quite nicely without being its own independent thing like the Assistants API component.

maybe what you want will be somewhat addressed by being able to serve LC agents via API, for any app, then in theory you could use them “externally” provided by another (API only) LC instance?

[danny-avila]

Other than that, nothing is stopping you from using some agentic process as a custom endpoint.

[Lucas-almma]

Danny - Are you referring to OpenAI's deprecation of Assistants, or you are announing LC Assistants will be dreprecated?

[danny-avila]

Both, more info: #6364

[mondo192]

Hi @danny-avila

Can you clarify how the upcoming public cloud hosting will affect the open source self-hosted version? Will any features be restricted or require payment, like the paid API for the code interpreter? Also, will there be any changes to the licensing of the open source version?

Thanks.

[danny-avila]

Good questions!

Basically, the cloud-hosted version will just be the same open-source app you can already self-host. We're not planning to take away or restrict any features from the self-hosted version because of the cloud offering. Everything you can do now for free when self-hosting will stay free. All planned features on the roadmap are planned to be open-source.

The only exception is the Code Interpreter API service – that's already a premium thing, and it'll stay that way.

The cloud version might eventually have some paid tiers (maybe for extra storage, resources, managed stuff, etc.), but the features will mirror what's available in the project repository.

And nope, no plans to change the open-source license of LibreChat (MIT License).

[stenebenau]

Hi Danny,

I'm using librechat within my company as the main chat applications for my employees (50). Everyone is really happy with it! However there are some features listed on the roadmap which makes it even better.

Admin features
Folders in chat & agents
Agent Sharing with groups
Memory
Deepresearch

Can you give an update on your roadmap, where you stand and where you can use some help? If needed we can invest some time back in the project?

Thank,
Sten

[danny-avila]

Hi @stenebenau these are all already being worked on:

• Admin features
• Agent Sharing with groups
• Memory (completed)

For the following, you would need to present a detailed user story of how you would implement it, before actually starting, and we can collaborate on the planning process then move to development. Mainly because these can be implemented in various ways, and what I think of when I see these words could be very different from what you are imagining.

• Folders in chat & agents
• Deep Research

On Deep Research, the Web Search tool was inspired by multiple papers on the topic, so when I imagine "Deep research" as a feature, it's using the same building blocks of the web search tool, with perhaps a more programmatic workflow and running for more than one pass.

[danny-avila]

Feel free to open new discussion posts regarding them, or get the conversation started on Discord.

[michnovka]

Hi @danny-avila , you mention memory is completed. I cant find this anywhere in docs, only an open issue - #5494 - is this not released yet, or did you mean this is completed via MCP? thanks for clarification!

[danny-avila]

Merged on the dev branch: #7760

[Defiect]

Has there been any further discussion on plans for a Deep Research feature?
In my opinion, even a mildly configurable Deep Research tool with sane defaults would be a killer feature (and is on the platforms already offering it). Nobody asked, but this is what I would want to see (and possibly work on) when it comes to this feature:

For one, I particularly think that the way Anthropic does theirs - using a multiple dispatch approach/mixture of agents approach has the most potential as a starting framework. (Claude opus will, for example, identify independent facets of the topic and deploy a sub-agent for each one)

After each sub-agent returns with their findings, the orchestration agent uses their findings to update its context on the original query and assesses any new facets that surfaced, along with identifying blind spots.

The next iteration of sub-agents are then sent to research those - a handful of these iterations later, the orchestration agent decides to end the research phase (either due to guiderails, awareness of token limitations, or via user-defined hard cap on iterations), and spends the remainder of the process synthesizing the report to return for the user.

This kind of Agentic research framework is not only much more effective than a single-agent "Web Search but long" implementation, it also doesn't require inventing a new wheel - agentic coding with sub-agents is already well established, and has basically identical programmatic structure.

TL;DR: bump, +1 for deep research

[hiddenadhd]

Please share your feedback and thoughts on the 2025 roadmap for LibreChat 🎉

https://www.librechat.ai/blog/2025-02-20_2025_roadmap

@danny-avila Would be great if someone updated the website home page with 2025 info and links. For instance, roadmap link in footer still goes to 2024 roadmap. And all stats/awards on homepage are from early 2024, making it seem like this project has died, which it clearly hasn't.

[berry-13]

updating the roadmap now, the stats are actually up to date. the badge was added late 2024, so it's not even that old yet

[petervcook]

@hiddenadhd Last week I submitted a PR to the docs to fix this link, it was accepted and merged in the day after your post. LibreChat-AI/librechat.ai#343 (comment)

https://www.librechat.ai/docs

[kivinju]

@danny-avila Thank you for the excellent work on this project! Since we're now in mid-July, I wanted to check on the current status of the Admin Panel. Do you have an updated timeline for when it will be available? I'm very interested in being among the first users to test it out.

[danny-avila]

It's taking longer than anticipated but is being actively worked on. Please be patient as this is our number one priority and requires a lot of architectural changes.

[gerkeac]

Thank you for all your hard work!

[pascalmariany]

Good job, @danny-avila 💪🏼 Hope you are doing well and enjoying the summer. Would love to test the admin panel in august.

[Zazi77]

@danny-avila as an experienced product minded dev, I applaud your focus on thought out architecture (and as such quality) over rushed deadlines.

Lack of quality will always haunt you.

As others, I anxiously await what is comming.

[baqianxin]

Hi @danny-avila , Thank you and the team for your contributions,I want to know the latest progress of the Admin Panel. I have been continuously paying attention to this part, and also the access of the custom OCR service.

[yaboijbigs]

Hey @danny-avila -- loving librechat. Do you have an update on the admin panel? Our team is really excited for it.

[danny-avila]

This is my latest update: #5960 (reply in thread)

I will give another update when the situation changes.

[a13782425]

Are there any plans to add more notebooks or knowledge bases?