← Back to all sparks
K

Koyeb

INFRA · APIS
Velocity2.1

Serverless platform for deploying full-stack apps and APIs globally with auto-scaling.

Koyeb signed a definitive agreement with Mistral while doubling down on Sandboxes for AI agents.

serverless infraai agent runtimegpu pricingephemeral sandboxesmistralscale-to-zero
Current state
Koyeb is a serverless platform that's leaning hard into AI infrastructure: GPU instances across A100/H100/L40S/RTX Pro 6000 with aggressive price cuts, the Sandboxes product (ephemeral isolated microVMs) now in public preview specifically for orchestrating AI-generated code, and a steady cadence of tutorials integrating Claude Agent SDK, GitHub Copilot CLI, OpenAI Apps SDK, Mistral Vibe, and Ollama. The platform-side fundamentals keep advancing — manual scaling, faster time-to-healthy deployments, scale-to-zero refinements, MFA/passkeys, partial-update PATCH endpoints. The April announcement of a definitive agreement with Mistral overshadows everything else.
Where it's heading
Koyeb is positioning as the runtime layer for the AI-agent economy: cheap GPUs, isolated sandboxes for arbitrary code execution, integrations with every major agent SDK. The Mistral agreement is the strategic capstone — exact terms are unclear from this feed, but a definitive agreement with a leading European model lab points to a vertically-integrated stack from model to runtime. Either Mistral is acquiring Koyeb's infra or vice versa; either way the AI-infra story tightens.
Prediction
Expect Mistral-specific integrations to land fast (one-click model deploy, native Mistral inference endpoints, possibly preferential pricing for Mistral workloads). Sandboxes will likely move from public preview to GA, and the agent-SDK tutorial cadence will continue as the primary discovery channel for AI-developer customers.

Recent moves

  1. 2mo ago

    Koyeb signs definitive agreement with Mistral

    ⚡ SPARK

    Koyeb announces a definitive agreement with Mistral, alongside manual scaling rolling out and Sandboxes accessibility via the Koyeb CLI. The Mistral deal is the headline — defining where Koyeb fits in the AI-infrastructure stack going forward.

    View source ↗
  2. 2mo ago

    Google Authentication Support for Koyeb Control Panel, Service Lifecycle Deletion Policy, and more

    Adds Google Authentication for the Koyeb control panel and a Service Lifecycle Deletion Policy. Routine governance and account-control depth on top of the recent MFA/passkeys work — a steady push to make Koyeb procurement-ready.

    View source ↗
  3. 2mo ago

    March 19, 2024New deployment events, improved deployment time after build succeeded, and moreNew deployment eventsImproved deployment tim…

    Entry appears to be a CLI reference documentation page captured by the changelog feed rather than a release announcement. The title mentions 2024 deployment events that don't match the 2026 publish date — best read as feed noise rather than substantive product change.

    View source ↗
  4. 3mo ago

    Partial Updates Now Valid Using the PATCH Endpoint of the Koyeb APINew Tutorial: Use Ollama to Test Multiple Code Generation Models with…

    PATCH endpoints now accept partial updates rather than requiring full payloads — a long-standing API papercut resolved. Paired with new tutorials on Ollama-driven code generation in Sandboxes, reinforcing the agent-runtime positioning.

    View source ↗
  5. 3mo ago

    January 16, 2026New Serverless GPUs, Auto-deletion and Lifecycle Management for Koyeb Sandboxes, and moreNew Serverless GPUs: RTX Pro 600…

    Adds RTX Pro 6000 to the Serverless GPU lineup and brings auto-deletion plus lifecycle management to Sandboxes. The Sandbox lifecycle work matters more than it looks — agentic workloads create a lot of ephemeral environments and the cost of forgetting to clean them up is real.

    View source ↗
  6. 3mo ago

    Multi-factor Authentication (MFA) and Passkeys now available for Koyeb accountsNew range for idle period for Instances using Scale-to-Zer…

    MFA and passkeys land for Koyeb accounts, alongside an extended idle range for Scale-to-Zero. Standard enterprise-readiness depth plus a meaningful tweak to the cost-control story for variable workloads.

    View source ↗