Context management and "checkpoints"

[oak moon]

One thing that really makes Warp unusable for proper AI coding IMO is the lack of context management and "checkpoints" that would allow bringing back the state before the specific message. I know and use git but doing it every message is both crazy and doesn't make sense. It does make sense when you code manually but when you code with AI and you give it access to change your files it is a different story.
Two main problems with using git manually while coding with Warp:

1. If i do it in the same conversation I clutter the context and the converstaion
2. Doing changes that the AI is not aware of like bringing back the previous state from git while continuing conversation with context after the changes were made confuses the models which leads to further problems.

I had it happened multiple time when I work on something, i do small incremental changes between commits so that it doesnt brake anything and then AI makes a mistake out of the blue on a simple task, reverting it back a few tasks leads first of all to more work to be done, second of all more credits taken, third broken context.

I'm begging you please implement this feature. Manuall message deletion and checkpoints would make the warp worth paying for even more and actually using it. I don't want to be rude but it for sure is more important then UI zoom 😛

[honest lava]

Hi @oak moon. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

---

You are absolutely not alone in your frustration—context management and the lack of robust "checkpoint" or undo features are among the most requested improvements from Warp users working with AI coding agents.

Current State:

• Warp does not yet have a native, user-facing checkpoint or message-level undo feature. Many users, like yourself, find that relying solely on git for every change is impractical and can break the conversational context with the AI, leading to confusion and workflow disruption. This is a well-documented pain point, especially when the AI makes a mistake and reverting via git causes the agent to lose track of the project state or context, resulting in further errors and wasted credits AI Checkpoints / Undo.
• Users have repeatedly requested features such as the ability to restore to a previous checkpoint, undo specific AI actions, or manually delete messages to clean up or correct the context. These requests are echoed in multiple issues and discussions, with many users stating that such features would make Warp much more usable and worth paying for AI Checkpoints / Undo, restore checkpoint.

What’s on the Roadmap:

• The Warp team has acknowledged these requests and confirmed that a checkpoint/undo feature is on their roadmap. For now, they recommend using source control (like git) to revert changes, but they recognize this is not an ideal solution for AI-driven workflows. Users are encouraged to upvote and subscribe to the relevant feature requests to help prioritize development AI Checkpoints / Undo.
• There is also ongoing discussion about improving context management in general, including better session persistence, conversation forking, and restoring previous conversations. Some of these features (like conversation history and forking) have already been implemented, but true checkpointing and granular undo are still pending Store/Restore Agent Conversation history, Warp AI history search.

Summary:

• You are correct: manual git usage is not a substitute for proper AI-aware checkpoints.
• The Warp team is aware and has this feature on their roadmap, but it is not available yet.
• For now, you can use conversation history and forking, but not true checkpoints or message-level undos.

If this feature is critical for your workflow, consider upvoting and commenting on the open feature request to help signal its importance to the Warp team: AI Checkpoints / Undo.

If you need more details on current context management workarounds or want to know how to use conversation history/forking, let me know!

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon: