Snabb.Back to overview

IMPLEMENTATION MAP · 1 SEPTEMBER 2026

The product, without wishful labels.

The catalogue is the destination. This page separates what is operating in the private Lab, what needs tightening before opening access, and what should deliberately wait.

Available Partial / qualifying Later

Available in Lab

Run

The load-bearing execution path is real today.
  • Firecracker-only customer workloads
  • Node, Python, Go, Rust and C/C++ runtime images
  • CPU, RAM, disk, IOPS, timeout and worker-aware admission
  • Live event ladder, logs, cancellation, reports and artifacts
  • CLI and VS Code entry points
  • Reusable dependency caches and immutable OCI builds

Available · qualifying

Workspaces

Persistent development is implemented; HA and resume behavior remains under active qualification.
  • Encrypted workspace generations
  • Multiple audited terminal sessions
  • Quick pause and local full-snapshot resume
  • Portable object-storage checkpoint fallback
  • Cross-image migration and generation restore
  • Command and file-operation audit without secret values

Partial

Prove

Evidence primitives exist; the full deterministic investigation UX is not complete.
  • Repeat and replay executions
  • Failure signatures and structured reports
  • Digest verification and checkpoint provenance
  • Pipeline version diffs and immutable activation
  • Still needed: guided Compare A/B, Causal and full bisect verdict pages

Available in Lab

Ship

Build and preview foundations are operational on isolated workers.
  • Versioned DAG pipelines and Git triggers
  • Harbor OCI publishing by verified digest
  • Trivy scan reports and artifact retention
  • Firecracker preview runtimes
  • Authenticated HTTP/WebSocket preview gateway
  • Still needed: public beta domain and polished release promotion flow

Partial

Agents

Bounded agent execution exists; supervision UX needs the new Snabb treatment.
  • Policy-bound execution and per-action identity
  • Plan approval and external-action approvals
  • Async command/file/network audit
  • Budget and runtime ceilings
  • Still needed: richer plan editing, agent templates and comparative outcome views

Available · one blocker

Govern

The policy and evidence chain is strong, with one operational launch blocker.
  • Team/project roles and scoped permissions
  • Immutable policy snapshots per execution
  • Network profiles, secrets broker and metadata blocking
  • Append-only audit ledger and export worker
  • Budget enforcement, exact metering and admin adjustments
  • Launch blocker: configure and qualify an external Object Lock/WORM destination

ADJUST NOW

Before a public beta

  1. Adopt the Snabb shell on authenticated pages without changing the proven BFF contracts.
  2. Finish workspace/session HA qualification through rolling gateway and worker changes.
  3. Complete deterministic investigation views: Repeat, Compare, Bisect and verdict evidence.
  4. Configure external WORM audit export and perform restore/verification drills.
  5. Close onboarding, empty/error/mobile states and accessibility acceptance against the design book.
  6. Make status, limits, reservation and refusal explanations consistent on every action.

MAKE LATER

After core reliability is measured

  1. SAML/SCIM enterprise administration and Sovereign/air-gapped deployment.
  2. Multi-region execution and data-residency selection.
  3. Broad third-party notification and issue-tracker integrations.
  4. Advanced causal inference and cross-project evidence federation.
  5. Public marketplace for runtime, pipeline and agent templates.

Use the working platform now.

The current private Lab remains the source of truth for actual capabilities while the Snabb interface is introduced page by page.

Open private lab ↗