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
- Announced October 9, Managed Deployments remains public preview. Free cannot run it; PAYG and Enterprise allow 3 and 10 simultaneous organization-level deployments.
- Public and private repositories on github.com are supported, with the Mistral GitHub App installed and access for the current account.
- Active means the worker registered workflows, not that an execution completed or produced correct output.
- New commits need explicit redeployment. Creation or startup beyond quota returns 403; stopping or deleting unused deployments releases slots.
Who can use it, and what do 3 and 10 limit?
Mistral announced Managed Deployments October 9, 2026 to host workflow workers, the programs executing tasks. It is public preview with potentially changing features and limits, aimed at existing Workflows developers reducing server maintenance.
Eligibility requires paid Pay-as-you-go or Enterprise; Free has zero deployment allowance. The announcement and guides do not specify hosting rates or regions. Paid eligibility is not free hosting; confirm billing and available regions before adoption.
Limits count simultaneously running managed deployments per organization: 3 for PAYG and 10 for Enterprise. A deployment is a separate worker instance. These numbers do not state task concurrency within a workflow or model-request, token, or thread quotas.
What does Mistral do after code is on GitHub?
Mistral Cloud clones the repository, builds an image from its Dockerfile, and starts the worker container. A Dockerfile specifies packaging and startup. You avoid preparing and maintaining worker servers, but the project still needs valid startup configuration.
Official onboarding requires Python 3.12 or newer, uv, and permission to create a GitHub repository. The Workflows generator provides an example and Dockerfile. Before pushing, follow guidance to confirm workflow registration and ensure local secrets are absent from commits.
Support currently means the public github.com service, not necessarily public repositories: repositories can be public or private. GitLab and GitHub Enterprise are future support, so do not apply this route to them yet.
If a private repository is absent, where do you check?
In Mistral GitHub App configuration, select the organization and intended repository and confirm App access. An organization owner installs or authorizes it if you are not an administrator. Private visibility is not itself an obstacle.
Then add connection credentials in Studio's GitHub App settings. Deployment lists show repositories with the App installed that the current account can access. Check App authorization and account access separately; making the entire repository public is unnecessary.
Does Active establish a successful workflow?
Select a repository, branch, and unique deployment name. Official guidance warns that same-name deployments take execution tasks from each other. After creation returns, cloning, building, and startup still follow, so a returned object is not completion.
Wait for Active on the deployment page and inspect registered workflows. Active establishes worker registration. For build failures, read deployment details, errors, and logs and check Docker build or startup rather than assuming the model is unavailable.
Next, follow the official example workflow and inspect its execution record and output separately. The guide provides waiting and output retrieval. Ended execution and business-acceptable answers are different; assess the latter against your own criteria.
Why must you redeploy after changing code?
Pushing changes to the deployed branch still requires explicit redeployment. Mistral resolves the latest commit, rebuilds the image, and restarts the worker. The guide does not promise automatic publishing after every push; a successful commit does not prove the running program changed.
Recheck deployment state, workflow registration, and new-version output. The worker receives a platform-injected service-account token without its own manually created API key. Deployment-management and workflow-calling clients still use API credentials; the whole process is not authentication-free.
How can you leave quota for testing?
Unexecuted hypothetical example: a PAYG organization with one production and two test deployments uses all three slots. Even with no current tasks, a fourth creation or startup returns 403 for deployment quota, not requests-per-second exhaustion.
Before adding a test environment, identify an unused deployment and stop or delete it to release a slot, then create or start the intended deployment. Distinct production and test names help avoid accidentally stopping an active worker.
Sources
Frequently Asked Questions
Can I select CPU or memory or view per-deployment metrics?
Current documentation says Mistral Cloud chooses instance size and every managed worker uses the same size. Per-deployment CPU and memory metrics are not exposed, so deployment count cannot determine computing power or throughput.
Can an organization request more than 3 or 10 deployments?
Mistral permits organizations to request a higher limit through support, without guaranteed approval. Until approved, plan production and tests within the current limit.
Can external-service credentials be stored in the repository?
Do not commit secrets. Onboarding directs external credential configuration to Secrets documentation. Injected worker identity does not automatically grant external-service access; follow the applicable configuration documents.