Skip to main content

NCT Tuning Studio

Upload your ECU .bin file, match it against the NCT demo library, and let the AI bridge apply verified recipes — then download the tuned file ready for your tool.

Version-guarded writes MCP tool contract DeepSeek / LLM driven Ready-to-download output

1 · Upload your ECU file

Drop your .bin file here, click, or press Enter to browse

Supported: bin · hex · ori · mod · map · kp — max 64 MB

3 · Choose your tuning intent

Pick a preset — the AI bridge will only apply changes backed by a verified demo recipe for your matched ECU. Anything unverified is refused honestly.

This is appended to the preset as an extra instruction for the AI bridge. Leave empty to use the preset as-is.

4 · AI tuning session

// session starting…

5 · Result

Download tuned file Original file Full audit report

NCT Demo Library

The verified demos the AI bridge can act on. A file that matches a stock demo exactly can have its verified recipe applied; everything else is read-only.

Datalog VE/AFR correction — $0, no LLM

Paste the same steady-state datalog analysis the open-source local tuning pipeline uses. Upload a CSV with rpm, map, afr_actual/afr, afr_target/target_afr (optional tps_dot for the steady-state filter) — you get a downloadable RPM×MAP VE-correction table.

Drop a datalog CSV here, click, or press Enter to browse

Headers: rpm, map, afr_actual, afr_target[, tps_dot]

What does this do?
  1. Steady-state filter — keeps only rows where |TPS change rate| < 5, so closed-throttle and tip-in transients don't pollute the AFR ratio.
  2. Ratio — AFR_actual ÷ AFR_target per row; >1 means the engine runs leaner than target (needs more VE), <1 means richer (needs less VE).
  3. Binning — groups by RPM (250 rpm bins) × MAP (5 kPa bins), exactly like the pandas groupby snippet.
  4. Correction — mean corrective multiplier per bin, downloadable as CSV to apply to your VE/AFR map.

Pending proposals

The AI bridge can propose byte changes, but never writes without your confirmation. Review each proposal's diff preview, then Apply (version-guarded — any from-byte mismatch refuses the whole patch) or Reject. Proposals expire after 48 h.

No session uploaded yet — proposals appear after the AI submits one.

ECU knowledge base

Metadata-only reference (patterns, not offsets) that grounds the AI bridge. Click a family to browse its tuning-relevant signatures — the AI reads these via ecu_lookup + find_signature, then locates actual bytes with find_pattern. Offsets are never invented.

Loading knowledge base…