Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Pando has served OpenAI's production environment since April 2026 and was built with Codex in ten weeks.
- OpenAI reports about 80% lower median push latency and nearly one petabyte of daily Git traffic; these figures describe internal workloads.
- It preserves ordinary Git client workflows. Different branches can be processed in parallel while updates to the same ref remain ordered.
- Stored packs and regional caches reduce some packing work at the cost of duplicate transfers in some cases; external eligibility, regions, prices and dates have not been announced.
First, what is Pando and can you use it today?
OpenAI introduced Pando in an engineering article dated November 2, 2026. It serves the company's internal monorepo: a single repository containing multiple projects. Pando entered production in April 2026, and the team built it with Codex in ten weeks. The public description provides no external signup, connection or trial process.
OpenAI's engineering article says the company is exploring ways to share the technology, but it gives no external eligibility rules, supported regions, prices, application path or release date. There is therefore no public basis for saying you can use Pando with your own Git repository today. An internal demonstration or exploratory statement should not be read as an open service.
Read the performance figures in context
OpenAI reports that, in its internal use, Pando reduced median push latency by about 80% and handles nearly one petabyte of Git traffic per day. These figures describe OpenAI's own workload and backend. They are not results measured across outside customers and do not mean every repository will become faster by the same amount.
The nearly one petabyte figure is the company's daily internal Git traffic, not a usage allowance for an individual user. To evaluate your own repository, measure it with your current host, repository size, region and team concurrency. Pando's internal figures cannot establish your migration benefit or personal quota.
Branches can run in parallel while updates to one ref stay ordered
Pando preserves ordinary Git client and protocol workflows while allowing different branches to proceed in parallel. Updates to the same ref remain ordered. A ref points to a commit; a branch name is a familiar example. The design reduces the need for every branch to share one global queue while retaining the order of updates to a single ref.
Different branches also do not have to share one snapshot in time; their repository states may represent different moments. This describes how Pando supports collaboration in OpenAI's monorepo. It does not change Git's branch semantics or show that an outside Git host is already using this service.
Packs and regional caches trade packing work against transfer
Pando uses stored Git packs and regional caches to reduce repeated packing work. A pack is a file containing bundled Git objects. Reusing existing data reduces some server computation for later requests.
The trade-off is that some situations may transfer extra duplicate data in order to do less packing work. This does not mean every clone or fetch always transfers less data, and a public Git client update should not be assumed to provide the caching behavior of Pando's server.
What can developers do now?
First, separate Git upstream improvements from the Pando service. OpenAI mentions public contributions to Git, but installing a Git client containing those improvements does not connect your remote repository to Pando. Check the actual release notes for each patch or client version.
If slow pushes or fetches are your problem, record your current host, repository size, operation times and team concurrency before deciding where the bottleneck lies. For example, repeating the same operation on the same repository at different times gives you a more useful starting point than applying another company's speedup figure. This is a measurement suggestion, not a result tested by PuppyIP.
Sources
Frequently Asked Questions
Are the interactive simulations in the official article production benchmarks?
No. They are synthetic teaching examples that explain the design. Treat them separately from OpenAI's reported internal production latency and traffic figures; they cannot predict your repository's performance.
Who wrote the engineering article, and how does Tibo's post relate to it?
The OpenAI article is credited to Aaron Friel, Taylor Blau and Ted Nyman. Tibo quoted the OpenAI Developers introduction and credited the team. Attribute technical details to the engineering article; his post does not announce external access or a quota reset.