VedaCODE architecture

Client access, control plane, data stores and GPU runtime.

VedaCODE Updated 1 min read

VedaCODE is a self-hosted stack installed by a single setup script.

TierComponents
Client access (HTTPS)Developer CLI (vedacode), VS Code extension, VedaCODE Studio web console
Control planeFastAPI on Ubuntu: API engine, build lifecycle, complexity router, WAF and guardrails
Data and runtimePostgreSQL 16 + pgvector (code search), Redis 7 (queues, cache), Ollama (16k context), Bubblewrap (sandboxed builds)
ComputeLocal NVIDIA GPUs (24 GB+ VRAM recommended); optional remote GPU workers (vedacode-worker)

Build lifecycle

  1. Planning — the plan records acceptance criteria, assumptions and open questions.
  2. Approval — four-eyes sign-off; execution is bound to the SHA-256 hash of the approved plan.
  3. Sandbox build — edits and commands run in Bubblewrap.
  4. Validation — unit tests and linters.
  5. Auto-repair — N-round patching.
  6. Completed — the verified diff is applied; otherwise the build ends as FAILED with full logs.