Cloud agent · active
Codex
OpenAI’s coding agent across ChatGPT, desktop, editor, terminal, worktrees, and cloud environments.
Visit Codex ↗Checked Oct 04, 2026 · review due Oct 08, 2026
- Best for
- Developers who want parallel local and cloud agent workflows connected through one ChatGPT account.
- Avoid if
- You need a self-hosted version of Codex cloud or every product surface to be open source.
01 / Editorial view
Where it fits
Strengths
- Local and cloud surfaces
- Built-in parallel worktrees and cloud environments
- Scheduled background work
Limitations
- Local model execution is available through the CLI; Codex cloud is not self-hosted
- Model availability depends on plan, client, workspace settings, and rollout
- Cloud features require ChatGPT authentication rather than API-key access
02 / Workflow
Surfaces and capabilities
- Autocomplete
- No
- Chat
- Yes
- Agent
- Yes
- Background agent
- Yes
- Review
- Yes
- Testing
- Yes
Editors
VS Code, Cursor, Windsurf, Xcode, JetBrains IDEs
Operating systems
macOS, Windows, Linux
Repository hosts
GitHub, GitLab
Languages
Language-agnostic
Bring your own key
Yes
Bring your own model
Yes
03 / Orchestration
Coordination model
Supported agents
Codex
Isolation
Git worktrees, Cloud environments
Task sources
ChatGPT, Editor, Terminal, Scheduled work
Maximum parallelism
Unknown
- Planning
- Yes
- Scheduling
- Yes
- Agent communication
- No
- Review gates
- Yes
- CI recovery
- Yes
- Merge workflow
- Unknown
- Session persistence
- Yes
- Observability
- Yes
- Human approval
- Yes
04 / Cost and deployment
What it costs to run
Individual
Free costs $0 and Go costs $8/month, with limited desktop access; Plus costs $20/month and Pro starts at $100/month.
Team
Business costs $20 per user/month with annual billing or $25 monthly, with a minimum of two users.
Enterprise
Contact OpenAI.
API-key usage is billed separately and does not include Codex cloud features.
05 / Privacy
Data handling
- Code processing
- The CLI can use local models through Ollama or LM Studio, or send context to a configured external provider. Codex cloud runs in OpenAI-managed environments; processing policies depend on the selected provider.
- Indexing
- Unknown
- Retention
- ChatGPT-authenticated use follows workspace retention and residency settings. API-key use follows the API organization's retention and data-sharing settings. Other providers apply their own policies.
- Training policy
- OpenAI does not train on Business, Enterprise, Edu, or API data by default; personal-plan users control training through ChatGPT data controls. Other model providers have their own policies.
- Enterprise controls
- SAML SSO, RBAC, SCIM, Audit logs, Retention and residency controls
06 / Evidence
Primary sources
- Codex authentication and data-handling policies ↗Checked Oct 04, 2026privacy.retention
- Codex IDE extension ↗Checked Sep 08, 2026classification, compatibility, capabilities
- Codex cloud ↗Checked Sep 08, 2026classification, compatibility.repository_hosts, capabilities, orchestration, deployment
- Codex models ↗Checked Oct 04, 2026models
- Codex pricing and plan controls ↗Checked Oct 04, 2026pricing, privacy
- Codex open-source components ↗Checked Sep 08, 2026deployment
- OpenAI Codex product page ↗Checked Aug 31, 2026classification, capabilities, orchestration
- Codex custom model providers ↗Checked Oct 04, 2026models, privacy.code_processing
- Codex CLI local model providers ↗Checked Oct 04, 2026deployment.self_hosted, deployment.self_hosted_interfaces, models.byo_model, privacy.code_processing
Changelog
What changed
- Codex local models and Kandev agent support recheckedOct 04, 2026The October 4 review corrects Codex local-provider support, updates its model and pricing details, and expands Kandev's verified agent list.
- September review clarifies billing and deployment pathsSep 22, 2026The September 22 review clarifies Codex API-key billing, Tabnine inference charges, and the maintained OpenHands components to use in a new trial.
Next