What Is Cursor 3.4 Cloud Agent Environments? (May 2026)
What Is Cursor 3.4 Cloud Agent Environments? (May 2026)
Cursor 3.4 launched on May 13, 2026 with a major upgrade to cloud agent infrastructure: multi-repo environments, Dockerfile-based configuration, audit logs, scoped egress, and Bugbot Effort Levels for AI code review.
Last verified: May 15, 2026
TL;DR
| Feature | What it does |
|---|---|
| Multi-repo environments | One agent env can mount multiple repos |
| Dockerfile config | Env-as-code with build secrets + layer caching |
| Cursor-led setup | Flags missing creds, falls back to base image |
| Version history + audit logs | Per-environment governance |
| Scoped egress + per-env secrets | Lock down where agents can call out |
| Bugbot Effort Levels | Default / High / Custom intensity tiers |
The headline change
Before Cursor 3.4, cloud agents ran in single-repo sandboxes configured via a YAML environment file. That worked for one-repo tasks but broke for monorepo coordination, cross-repo refactors, or backend+frontend changes.
Cursor 3.4 swaps the YAML model for Dockerfile-based, multi-repo environments with explicit version history, audit logs, and per-environment secret scopes — making cloud agents usable in regulated team contexts.
What’s in 3.4
1. Multi-repo environments
A single cloud-agent environment can now mount and operate across multiple repos. Useful when:
- A feature spans frontend + backend repos.
- A monorepo conversion needs to read both the old and new layouts.
- An infra change touches both the app and the deploy repo.
2. Environment configuration as code (Dockerfile)
Configuration moves to standard Dockerfiles with:
- Build secrets support (no more leaking tokens into layers).
- Layer caching for fast rebuilds.
- Plain
FROM/RUN/COPYsyntax that any team already knows.
This replaces YAML for new environments.
3. Cursor-led environment setup
When you create a new env, Cursor:
- Inspects the repo and proposes a config.
- Flags missing credentials before the agent runs.
- Validates the build.
- Falls back to a Cursor-provided base image if your custom build fails.
4. Governance and security
Each environment now has:
- Version history — see every config change.
- Audit logs — who changed what, when.
- Scoped egress — limit which domains the env can call out to.
- Per-environment secrets — secrets scoped to one env, not the whole team.
5. Bugbot Effort Levels
Bugbot is Cursor’s AI code reviewer (exited beta in July 2025). 3.4 adds three tiers:
- Default — current balance of speed and depth.
- High — deeper analysis, more cross-file context, more thorough.
- Custom — team admins tune specific knobs (depth, models, patterns).
Admins set team defaults; users can override per repo or PR.
How 3.4 cloud envs compare
| Capability | Cursor 3.4 | Claude Code Cloud Agents | OpenAI Codex Cloud |
|---|---|---|---|
| Multi-repo | ✅ | 🟡 (single repo per session) | 🟡 (single repo) |
| Env-as-code | ✅ Dockerfile | 🟡 implicit | 🟡 implicit |
| Audit logs | ✅ | 🟡 limited | 🟡 limited |
| Scoped egress | ✅ | 🟡 | 🟡 |
| Effort tiers | ✅ Bugbot | ✅ Claude reasoning tiers | ✅ effort param |
| Governance UI | ✅ | 🟡 CLI-heavy | 🟡 web-only |
| Best for | Team workflows, regulated orgs | Autonomous long-running tasks | Single-repo automation |
Who Cursor 3.4 is for
✅ Teams with:
- Multi-repo product surfaces (web + iOS + backend).
- SOC 2 / regulated audit requirements.
- Already-standardized Dockerfile-based CI.
🟡 Single-developer / single-repo: 3.4 is fine but the new features mostly don’t matter — Claude Code or local Cursor work just as well.
❌ Air-gapped / fully offline: still need an on-prem path.
What 3.4 fixed
- The single-repo limitation that forced people to clone-and-symlink to fake monorepo work.
- The YAML config sprawl that made env configs hard to review.
- Stealthy egress — agents couldn’t be reliably constrained to approved domains.
- Bugbot one-size-fits-all — heavy users now get high-effort review; light users get faster default review.
Known issues at launch
- A community-forum post on May 14, 2026 reported “Cloud Agent from Plan Always Fails” in version 3.2.16; upgrading to 3.4 resolves it for most users.
- High-effort Bugbot can be noticeably slower on large PRs — use it for finals, not draft pushes.
- Multi-repo environments cost more compute credits per run than single-repo.
Pricing impact
Cursor 3.4 doesn’t change list pricing:
- Free — limited cloud-agent usage.
- Pro ($20/mo) — standard cloud-agent quota.
- Business — multi-repo and audit features fully available; admin governance UI.
- Enterprise — SAML/SSO, custom egress allowlists.
The multi-repo features are available across tiers but only really shine on Business/Enterprise where governance matters.
How to upgrade
- Update Cursor to 3.4 via Help → Check for Updates.
- Open any cloud-agent environment.
- Click “Migrate to Dockerfile config” or create a new env from scratch.
- In team settings, set the default Bugbot Effort Level for your org.
- Configure scoped egress allowlists for environments handling proprietary code.
Risks and watch-outs
- Dockerfile drift — env configs are now code; treat them like any other shared infra (PRs, reviews).
- Egress allowlist gaps — easy to over-allow at first; audit after a week.
- Bugbot High cost — quota burn is materially higher; set per-repo defaults thoughtfully.
What to watch next
- Cursor 3.5 is rumored to add agent-to-agent handoff in a multi-repo env.
- SpaceX/Colossus integration for Cursor’s own model training (post the April $60B option deal).
- Bugbot model upgrades — currently uses a tuned Claude/GPT mix; may shift after Cursor’s own model lands.
Related reading
- Cursor 3 Agents Window vs Claude Code Parallel Agents (May 2026)
- JetBrains AIR vs Cursor 3 vs Claude Code (May 2026)
- Codex Cloud vs Claude Code Cloud vs Cursor 3 Cloud (April 2026)
- SpaceX Cursor $60B Deal (April 2026)
Sources: Cursor changelog (cursor.com/changelog), Cursor blog (cursor.com/blog/cloud-agent-development-environments), developer-tech.com, ai-dev-blog.com — May 13–14, 2026.