Zectre

Zectre is an AI-native compute platform: run serverless or dedicated, and deploy AI agents, inference endpoints and data pipelines on infrastructure that adapts to the workload. You describe an app — a model, the compute it needs, how it should scale — and deploy() puts it live.

The platform covers six things. AI inference: models deployed to GPU-backed endpoints that scale from zero to thousands of concurrent requests, with vLLM, TGI and custom runtimes. Adaptive compute: serverless for bursty work or reserved dedicated instances for sustained throughput, chosen per service rather than per account. Microservices: independent services in isolated runtimes, with discovery, load balancing and zero-downtime rollouts built in. A data layer: managed Postgres, Redis and object storage, serverless or provisioned, with branches for development and replicas for production. Agent orchestration: models, tools and data sources chained into autonomous agents, with memory, tool calling and human-in-the-loop controls. And observability: traces, metrics and logs across every service and model call, with cost attributed per request and no third-party SDK.

Underneath is a Nomad-inspired orchestrator written in Rust, built first to run Tish native binaries. A control plane of zectre-api (the REST API), zectre-scheduler (job scheduling and allocation) and zectre-registry (artifact storage for compiled binaries) drives zectre-agent on each worker node, which executes and supervises the workloads, with zectre-proxy doing ingress, dynamic routing and load balancing in front. zectre-build is the CI service that compiles Tish source into those artifacts, and zectre-catalog handles GitHub OAuth, PKCE sign-in for the CLI and dashboard, JWKS and the cluster registry. Deploys can be driven from Git, through an OAuth app or a GitHub App with webhooks or pull-only access for private repos.

Clients are the zectre CLI and a web dashboard with one-click installs and an applications view, and a new project starts from npx create-zectre-app. The platform deploys to your own server in one command — compiling there, or cross-compiling Linux binaries on a Mac and uploading them.