Cursor 3.11 is worth learning if long agent conversations are becoming harder to navigate. Side Chats let you investigate a question without derailing the main task, conversation search makes old agent work recoverable, and redesigned project pickers reduce friction when you move between local and cloud work.
The headline feature is not a new model. It is better control over context. That matters because many expensive AI coding mistakes begin when a useful conversation becomes too broad: a quick question changes the plan, an old decision cannot be found, or an agent starts in the wrong repository.
This guide was verified against Cursor’s official 3.11 changelog and pricing page on July 13, 2026. The screenshots are from Cursor’s official release page. No benchmark results or unsupported hands-on claims are presented as fact.
Official Cursor 3.11 release page, dated July 10, 2026.
What is the quick verdict on Cursor 3.11?
Upgrade and change your workflow if you regularly do any of the following:
- Keep one agent chat open for hours or days
- Interrupt implementation to investigate a side question
- Search old chats for a decision, filename, or failed approach
- Switch between local, cloud, and remote projects
- Operate cloud agents with team hooks or observability
The most useful change for most developers is Side Chats. Conversation search is the feature that will save the most time over a month. The new cloud agent hooks are the most powerful change for platform teams, but they also deserve the most careful security review.
If you mostly use Cursor as an autocomplete tool and rarely open the agent, 3.11 will feel incremental. Our broader Cursor AI vibe coding guide is a better starting point for the core product.
What changed in Cursor 3.11?
| Feature | What changed | Who benefits most |
|---|---|---|
| Side Chats | Start a separate, durable agent conversation that inherits context from the main chat | Developers who investigate tangents during implementation |
| Conversation search | Search agent transcripts from the Agents Window with Cmd+K | Anyone with a large history of agent work |
| Current-chat search | Use Cmd+F with match navigation inside a conversation | Developers reviewing long transcripts |
| Project and repo pickers | Clearer scope for Computer, Cloud, or remote work; connect GitHub, GitLab, or Azure DevOps in the picker | Multi-repo and remote teams |
| Cloud agent hooks | New events around prompts, responses, thoughts, stops, and subagents | Platform, security, and observability teams |
Cursor calls this release “Side Chats and Conversation Search.” That is an accurate description: 3.11 is primarily a workflow-management release, not a model-quality announcement.
How do Side Chats work?
Cursor’s official changelog says you can open a side chat with /side, /btw, or the plus button at the top of the chat panel. The new chat inherits context from the main conversation, so you do not have to repeat the entire problem.
A Side Chat is not a temporary tooltip. Cursor describes it as a durable, full agent conversation that you can revisit and continue. By default it is designed to read, search, and answer, which makes it a natural place to investigate without changing the main implementation thread.
Use a Side Chat for questions that may change your confidence, not your task
Good Side Chat prompts include:
- “Where is this permission enforced, and what other routes depend on it?”
- “Explain why this test double no longer matches production behavior.”
- “Search for earlier migrations that changed the same column.”
- “Compare these two approaches without editing files yet.”
- “What assumptions in the current plan are not supported by the repository?”
Keep the main chat focused on the approved implementation. Use the Side Chat to gather evidence, challenge an assumption, or explain unfamiliar code.
Do not create a Side Chat for every small question
Too many parallel chats can recreate the same navigation problem in a different form. A useful rule is:
- Stay in the main chat when the answer directly advances the current step.
- Open a Side Chat when the answer could branch, contradict, or delay the current step.
- Start a new main task when the work deserves its own acceptance criteria and review.
This separation is similar to the way a developer uses a short research spike before committing to a larger change.
Why is inherited context useful and risky?
Inherited context saves setup time. The Side Chat can understand the current task, files, and prior discussion without a large copied prompt. But inherited context can also carry a wrong assumption into the side investigation.
Start important Side Chats with one sentence that fixes the scope:
Read and search only. Do not edit files. Verify whether the current assumption about authentication is supported by the repository.
That sentence gives the chat a clear job, an action boundary, and an evidence standard. It is more reliable than simply asking “Is this right?”
If your work requires deeper autonomous implementation rather than a context-preserving tangent, compare Cursor with the tools in Best AI Coding Tools 2026 and Cursor vs Windsurf vs Zed 2026.
How does conversation search work?
Cursor 3.11 adds two search paths:
- In the Agents Window, use
Cmd+Kto search agent transcripts. Cursor says it builds a local search index that scales to thousands of conversations. - Inside an existing conversation, use
Cmd+Fto find text, jump between matches, and see a match counter.
This distinction is practical. Cmd+K answers “Which conversation contained that decision?” Cmd+F answers “Where in this conversation did we discuss it?”
Official Cursor screenshot. Cmd+K searches across agent transcripts; Cmd+F searches within the open conversation.
Search for evidence, not vague topics
Queries such as “auth” or “database” will return too much. Search for durable artifacts:
- An exact error message
- A function, class, table, or environment variable name
- A ticket or pull request number
- A phrase from the acceptance criteria
- A command that failed
- The name of a migration or test
When you find a decision that will matter again, move it out of chat and into the repository: an issue, ADR, test, comment, or documentation file. Searchable chat history is useful, but it should not become the only record of architecture.
What changed in the project and repo picker?
Cursor redesigned project and repository selection so the scope is clearer. The official changelog describes separate context for This Computer, Cloud, or a specific remote machine. It also says you can create a project and connect GitHub, GitLab, or Azure DevOps without leaving the picker.
This is a small interface change with a large risk benefit. Starting an agent in the wrong repository or remote environment can waste time and, with write access, create incorrect changes.
Official Cursor screenshot. The picker now makes local, cloud, and remote scope more explicit.
Use a 10-second scope check before every agent task
Before submitting a task, verify:
- Repository name
- Current branch
- Local, cloud, or remote environment
- Whether more than one folder is selected
- Whether the connected provider is the intended account
- Whether the task should be read-only or allowed to edit
This check is more valuable than a clever prompt. Context mistakes are deterministic; better wording cannot rescue the wrong repository.
What are the new cloud agent hooks?
Cursor 3.11 adds hooks that observe or control the agent conversation itself. The official changelog names events including:
beforeSubmitPromptafterAgentResponseafterAgentThoughtstopsubagentStart
Cursor also describes hooks around prompts, responses, thinking, subagents, compaction, and turn completion. These complement existing team hooks around tool execution and file or shell work.
What can teams use them for?
Potential uses include:
- Recording task-level metrics without parsing the UI
- Enforcing a required prompt header for sensitive repositories
- Alerting when a subagent starts
- Tracking stop reasons and unfinished turns
- Adding a review step before certain workflows continue
- Connecting agent activity to internal observability
These are workflow examples, not a claim that Cursor ships every policy out of the box. Teams still need to implement, test, and maintain their hook logic.
Why do hooks need a security review?
Prompt and response events may contain source code, file paths, issue content, customer data, or secrets pasted by a user. Logging every event “for observability” can create a new sensitive-data store.
Before enabling hooks in a team environment, answer:
- What exact fields are collected?
- Are thoughts or responses stored, and for how long?
- Can secrets be redacted before transmission?
- Who can read the hook output?
- What happens when the hook endpoint fails?
- Can a hook block work, or is it observation-only?
- Is the data allowed to leave the development environment?
Start with metadata such as event name, timestamp, repository identifier, and success state. Add content only when there is a defined need and an approved retention policy.
What is the best 20-minute setup after upgrading?
Minutes 0-5: Test Side Chats on a read-only question
Open an active agent conversation. Use /side or /btw and ask the new chat to locate evidence without editing. Confirm that the inherited context is useful and that the side task remains separate.
Minutes 5-10: Search for an old decision
Open the Agents Window and use Cmd+K with an exact error, file, or ticket name. Open the result, then use Cmd+F inside the conversation. This confirms both search levels.
Minutes 10-15: Inspect project scope
Open the redesigned picker and identify how local, cloud, and remote contexts are labeled. Remove stale recent projects if they make selection ambiguous.
Minutes 15-20: Audit hooks before enabling them
If you use cloud agents, list existing hooks and owners. Do not turn on content logging as an experiment. Write down the fields, destination, retention period, and failure behavior first.
Save this checklist in your team’s onboarding notes. It is short enough to repeat after a major Cursor update.
Does Cursor 3.11 change pricing?
The 3.11 changelog does not announce a new plan structure. As of July 13, 2026, Cursor’s live pricing page lists:
- Hobby: free, with limited agent and Tab usage
- Pro: $20 per month for individuals, with extended agent limits and access to features including frontier models, MCP, skills, hooks, and cloud agents
- Teams: $40 per user per month, with team management and policy features
Included usage, model availability, and on-demand billing can change. Check Cursor’s live pricing page before upgrading or enabling paid overages.
The right cost question is not “Does Side Chat cost extra?” It is whether better context separation and search reduce repeated investigation and abandoned agent work enough to matter to your team.
Who should skip changing their workflow?
You do not need to reorganize everything around 3.11 if:
- You mostly use Tab completion and short agent prompts
- Your chat history is small and easy to navigate
- You work in one local repository
- Your team does not use cloud agents or hooks
- Important decisions already live in issues and repository documentation
Install the maintenance update, but do not create process for features you do not need.
If you are deciding whether Cursor itself is the right editor, read Cursor vs Windsurf vs Zed before optimizing a workflow inside the wrong tool.
Common mistakes after the update
Using Side Chats as untracked task branches
Side Chats are durable, but they should not replace issues, branches, or acceptance criteria. If the tangent becomes real work, promote it to a separate task.
Treating chat search as documentation
Search helps recover a decision. It does not make the decision visible to a future maintainer. Move durable knowledge into the repository.
Leaving stale projects in Recents
The redesigned picker is safer only if the list is understandable. Remove obsolete repositories and confirm scope before starting an agent.
Logging agent content without a data policy
Hooks can expose much more than ordinary operational metrics. Begin with minimal metadata and add content only after security and privacy review.
Comparing releases by feature count
Cursor 3.11 is valuable because it reduces context friction. Measure time to recover decisions, number of derailed tasks, and wrong-repository incidents instead of counting new buttons.
Frequently asked questions
How do I open a Side Chat in Cursor 3.11?
Use /side, /btw, or the plus button at the top of the chat panel. The new chat inherits context from the main chat and remains available for follow-up.
Does a Side Chat modify my code?
Cursor describes Side Chats as full agent conversations and says their default role is to read, search, and answer. Treat permissions as something to verify in your environment. For investigations, state “read and search only; do not edit files” in the prompt.
What is the difference between Cmd+K and Cmd+F?
In the Agents Window, Cmd+K searches across agent transcripts. Inside an open conversation, Cmd+F searches that conversation and lets you move between matches.
Is conversation search local?
Cursor says it builds a local search index that scales to thousands of conversations. Review the current product documentation and your organization’s data policy if storage location is a compliance concern.
Do the new hooks replace existing tool hooks?
No. Cursor presents them as additional hooks for the agent conversation lifecycle, alongside existing team hooks for tool execution and file or shell work.
Is Cursor 3.11 better than Codex or Claude Code?
This release improves Cursor’s editor workflow, especially context separation and retrieval. It does not establish a universal model or agent winner. Compare real repository tasks using our OpenAI Codex app guide and Claude Code guide.
Final recommendation
Adopt Cursor 3.11 in three steps:
- Use Side Chats for evidence-gathering tangents, while keeping implementation in the main task.
- Use
Cmd+Kto recover an old conversation andCmd+Fto inspect the chosen transcript. - Review repository scope and hook data before allowing a cloud agent to act.
The release is useful because it makes agent work easier to separate, recover, and govern. Those improvements are less flashy than a new model, but they address problems that appear every day in long AI-assisted coding sessions.
Official sources
