01
A pod, not a sandbox
Every project lives in a dedicated Kubernetes pod with a persistent volume. No shared filesystem. No shared exec. Your security team can review the blast radius on one page.
ProyectaEnterprise
Sign in →For engineering organizations
AI that shipsinside the codeyou already have.
Proyecta Enterprise connects your real repositories — data dumps, service meshes, internal auth and all — and builds features into them. Every session runs in an isolated Kubernetes pod scoped to your workspace.
Pilot workspace provisioned in under two minutes. No procurement paperwork to see it run.
Typical cold-start
5.4s
Languages
40+
Workspace isolation
Per-pod
Live · scroll to explore
The insight
Every AI coding tool skips the hardest part.
They optimize for the first thousand lines of a fresh project. At fifty engineers, that is the easy part. The hard part is the system you already have — the one with a polyglot monorepo, three databases, internal auth, and eight years of decisions encoded in config files.
We built Proyecta for the hard part.
02
Reads the infra you already declare
Postgres with your seed dump. Redis on your version. Private registries, internal OIDC, polyglot monorepos. We read proyecta.config.json and provision the full dependency graph alongside your code.
03
Ships as pull requests
Multi-step plans that survive restarts, compile cleanly, and land as a PR on a session branch. Protected branches stay protected. You review what ships.
Proof
From your repo URL to a shipped pull request.
Every Proyecta session walks these four phases. No special setup for the first one, no different flow for the fiftieth.
Phase 01
Connect a repository.
Install the Proyecta GitHub App on your organization. Pick a repo and a base branch. Proyecta never copies your code to our storage — it clones into an isolated pod for the session.
Phase 02
Read the real stack.
We parse proyecta.config.json. Services declared there become sidecars: Postgres with your seed dump, Redis on your version, whatever you already ship with.
Phase 03
Build in chat.
Describe the feature. The agent plans milestones, writes the code, compiles, runs tests, and iterates against the live services until the preview renders correctly.
Phase 04
Ship as a PR.
Every change lands on a session branch. Protected branches stay protected. A pull request opens on your side, with every tool call audited and timestamped.
acme / core-platform · runtime
isolated
git clone—github.com/acme/core-platform
detect config—postgres · redis · node 20
restore database—seed.tar.gz · 842MB
install deps—pnpm install · 1,284 pkgs
start services—api :3000 · worker :3001
agent working—waiting for prompt
pull request opened——
proyecta/add-audit-trail
7 files · +312 −14
Scroll to follow the session
Boundaries
Every control your security review will ask about.
Isolation, secret scope, and audit trails are not afterthoughts here. They are the shape of the platform.
SOC 2 Type II in progress · ISO 27001 in progress
Pod-level isolation
One Kubernetes pod per project with a dedicated 10Gi PVC. No shared exec, no cross-tenant filesystem.
Secrets with scope
Variables are tagged Pod, Convex, or Repo. VITE_-prefixed keys blocked from Pod scope at the API boundary — they would leak to the browser.
Protected-branch blocklist
main, master, and develop are off-limits. Every change lands on a session branch and opens a PR for your review.
Auditable tool calls
Every read, write, shell command, and git action the agent takes is logged with timestamps, scoped to the workspace.
Next to the alternatives
Honest column by column.
Partial support means the capability exists in some form but not as a primitive.
Capability
Proyecta
Copilot-class
Agentic IDEs
Runs against your real repository
Boots your Postgres with real data
Provisions infra sidecars from config
Isolated pod per project
Protected-branch blocklist
Persistent milestone trees
Self-hosted in your VPC
“We pointed it at a monorepo with eighty-four services and twelve years of history. The first working pull request opened in under three minutes. That changed the conversation internally.”
Answers
The questions your team is about to ask.
In your GitHub repository, under your GitHub App installation. Proyecta clones into an isolated Kubernetes pod for the duration of a session. The pod has a 10Gi PVC tied to your workspace, encrypted at rest. When you suspend a project, the pod is deleted and only the PVC remains.
No. Proyecta does not train any model on customer code. We use frontier models from Anthropic and OpenAI under enterprise contracts that disallow training on our customers’ data.
Yes, on Enterprise plans. We ship a Helm chart for the agent runtime, pod provisioner, and control plane. The AI server can run inside your VPC against your own model API keys.
A buildpack-based container with Node, Nix, and Mise preinstalled; your repository cloned in; optional infrastructure sidecars (Postgres, Redis, MongoDB, MySQL, MailHog) declared in proyecta.config.json; and a gRPC agent runtime binary that streams tool calls back to the control plane.
Cursor is a local IDE. Devin runs an agent on an ephemeral VM. Proyecta runs a persistent, isolated Kubernetes pod per project with your real infrastructure sidecars and a milestone-based planner that opens pull requests. The comparison table above is the short version.
Median 90 seconds for a repository under 1GB with standard infrastructure. Buildpack installs are cached per organization. Warm pods resume from suspended PVCs in under five seconds.
Show us your hardest repository.
Twenty minutes with a Proyecta engineer. Bring a real bug, a real feature ticket, or a branch you have not touched in months. We run it live.
Pilots start the same week · No procurement paperwork to see it run