UAI Thaladex docs
runbookuapiredacted view

UAPI Grok Autorouter Friendly Route Next Plan 2026-06-22

Grok candidate friendly route plan for NUC, Cortex memory, and Colab/cloud GPU burst lanes.

Updated 2026-06-22·Freshness: Reference·Sensitive tokens or credentials were hidden before display.

UAPI Grok Autorouter Friendly Route Next Plan 2026-06-22

Grok candidate friendly route plan for NUC, Cortex memory, and Colab/cloud GPU burst lanes.

Promotion Metadata

  • Source feedback id: 114
  • Source feedback ids: 114
  • Promotion request id: 45
  • Feedback category: beneficial
  • Feedback source: [REDACTED:high-entropy]
  • Feedback created at: 2026-06-22T18:22:35.145173+00:00
  • Target slug: [REDACTED:high-entropy]
  • Review state: review
  • Reviewer note: AI lane auto-promoted redacted feedback into internal catalog context.

Source Feedback Body

Feedback 114

  • Product: uapi
  • Category: beneficial
  • Related slug: [REDACTED:high-entropy]
  • Source: [REDACTED:high-entropy]
  • Created at: 2026-06-22T18:22:35.145173+00:00

Grok Autorouter-Friendly Route Next Plan (2026-06-22)

ALIVE: true
Lane: Grok / UAPI hardware autorouter (NUC 140T/NPU)
Sources: 2026-06-22 [REDACTED:high-entropy].md, [REDACTED:high-entropy].md, grok-benchmark-preflight-command-plan.md, [REDACTED:high-entropy].md, agent-work-queue-v0.md.

Verified Facts (candidate only)

  • NUC-IPEX: Core Ultra 9 285H (16c), ~64 GiB RAM, Arc 140T (drv 32.0.101.8331), NPU AI Boost (OK), boot NVMe WD_BLACK SN8100 2TB (healthy). SN850X: not present.
  • Python 3.13.12 default: no [REDACTED:high-entropy].
  • WSL absent. Idle OK. C: ~704 GiB free.
  • Prior blocker (sanitized inventory) resolved. Current blockers: operator gate, Codex contracts, stacks absent, bench evidence.
  • Candidate routes must remain candidate-only until small Gemma/SLM bench on target devices + re-bootstrap.

Candidate Routes (names only; no claims)

  • local.nuc.gemma.npu : NPU (AI Boost) lane for low-power small Gemma/SLM.
  • local.nuc.gemma.140t : Arc 140T iGPU lane for higher-perf Gemma.
  • local.nuc.cpu.reserved : Reserved cores on 285H (policy decision pending; fallback).
  • lab.cortex.memory : Lab Cortex memory/knowledge hybrid lane (candidate).
  • cloud.colab.gpu.burst : Transient Colab / cloud GPU overflow burst lane (candidate).

Gate Plan (preflight complete; next is gated)

  1. Await explicit operator gate for install + small Gemma (2B-class or equiv SLM, 1-3 short prompts).
  2. On gate: read-only preflight cmds (reverse path), clean env check, then approved pip + export + 1-shot bench.
  3. Capture: first_token_ms, tokens_per_sec, load_time_s, peak_mem_mb, device, stability, fallback. Sanitize results.
  4. Re-bootstrap NUC state post-test.
  5. Promote evidence; only then consider route activation in PlacementDecision.

Unresolved Separate Jobs (do not conflate)

  • cloud.colab.gpu.burst lane: independent burst capacity; requires separate Colab auth/config + cost/queue handling. Not tied to NUC.
  • SN850X re-inventory: current NUC uses SN8100 (boot/system). SN850X location/status unknown; treat as distinct hardware inventory task. No SN850X-based route here.

Next Safe Artifacts / Handoff

  • Await Codex promotion of preflight evidence + this review.
  • Operator/Codex gate decision for limited Gemma smoke.
  • Update route matrix post-bench (no mutation now).
  • All actions: local diary only. No route changes, no installs, no inference.

Blockers: operator small-Gemma gate pending; Codex contracts (PlacementDecision, inventory, routes) not baselined; AI stacks absent on NUC; no key for any cloud; SN850X and Colab lanes separate.

Sources used: the five listed local diaries. Public research limited to cited OpenVINO/Intel primary repos for primary stack context only.

Report: ALIVE: true | Next: gate + bench evidence only after approval. Constraints: local artifacts, read-only authority, candidate names only. End.