Intelligent by Design

Not every issue needs the same resources. Nightgauge automatically adapts its approach — choosing the right model, the right depth of analysis, and the right pipeline path for every issue.

Right-Sized for Every Issue

Every issue is automatically assessed and categorized. The result drives how deeply the pipeline analyzes your codebase, which AI model is selected, and how the pipeline is routed.

XS
Trivial

Typo fixes, single-line changes, config tweaks

Fast-tracked through an optimized path — finished in minutes

S
Small

Simple bug fixes, documentation updates, small refactors

Streamlined pipeline with reduced overhead

M
Standard

New features, moderate refactors, API changes

Full pipeline with comprehensive analysis

L
Complex

Multi-file features, architectural changes, cross-cutting concerns

Extended analysis with deeper code understanding

XL
Extensive

Large-scale refactors, new subsystems, critical path changes

Maximum analysis depth with thorough exploration

5x
Faster for simple work. By adapting the pipeline to the actual complexity of each issue, Nightgauge avoids wasting time and resources on work that doesn't need it.

The Right Model for Every Task

Nightgauge picks the right AI model for every stage of every issue — and keeps tuning that choice with your own cost and outcome history, not benchmarks.

Cost Optimized

Formatting, config changes, and documentation tweaks run on fast, lightweight models. Deep reasoning is reserved for work that actually needs it — so you don't pay for reasoning power you don't need.

Automatically Matched

Every issue is scored for complexity and each stage is matched to a model tier — fast for simple work, balanced for standard work, maximum reasoning for the hard parts. Scoring is deterministic; tuning happens within your mode's envelope.

Learns Your Costs

The router tracks cost per success for every model and stage in your workspace. When a cheaper tier delivers comparable results, it quietly takes over — evidence, not guesswork.

Set the Envelope. Routing Fills It In.

A mode is an envelope, not a preset. It sets the floor and ceiling of the model range — adaptive routing still picks per stage inside that band.

Efficiency ≈ ½× cost
Haiku Sonnet Opus Fable 5

Capped at Sonnet, resting on Haiku wherever it suffices. Built for chores, docs, and high-volume runs where throughput matters more than depth.

Elevated Default baseline
Haiku Sonnet Opus Fable 5

The balanced default. The full adaptive range — Haiku for plumbing, Opus for hard reasoning — with nothing pinned.

Maximum ≈ 4× cost
Haiku Sonnet Opus Fable 5

Cost-no-object: every stage pinned to Opus at high effort. The one mode that doesn't route — it pins, deliberately.

Frontier premium · surgical
Haiku Sonnet Opus Fable 5

Unlocks frontier reasoning where it earns its rate: only complex planning and implementation escalate to Fable 5. Validation is capped at Opus, plumbing stays on Haiku, and trivial issues don't pay frontier prices.

Tuned by Your History

With cost-aware routing on (the default), the router compares each tier's cost per success across your own runs — a minimum sample size and success-rate floor keep it honest — and prefers the cheaper tier when results are comparable. Always inside the mode's envelope: never below the floor, never above the ceiling.

Graceful Under Quota

Frontier capacity is often metered separately from the rest of the model family. If Fable 5 hits a usage limit mid-run, the stage falls back to Opus and retries — the pipeline keeps moving instead of stalling. Only a genuine account-wide limit pauses for a quota cooldown.

Want surgical control? Custom mode pins any stage to any model — Auto, Haiku, Sonnet, Opus, or Fable 5 — right from the VS Code status bar.

Smart Pipeline Paths

Not every issue needs every stage. The system automatically determines the optimal pipeline path — streamlining simple work while giving complex issues the full treatment.

Fast Track Minutes

Simple changes like typo fixes, config updates, and documentation tweaks are fast-tracked through a streamlined path. No unnecessary overhead.

Standard Under an hour

Features, bug fixes, and moderate refactors get the full pipeline experience. Planning, implementation, validation, and delivery — all automated.

Deep Analysis Thorough

Large-scale changes get extended analysis with deeper code exploration. The pipeline spends more time understanding your codebase to get it right.

Predictable Where It Matters

Consistent Decisions

  • Complexity assessment
  • Pipeline routing
  • Error handling
  • Safety controls

Repeatable, predictable behavior you can rely on.

AI Where It Counts

  • Code generation
  • Bug analysis
  • Test creation
  • Implementation planning

Full AI reasoning for creative and complex tasks.

Knowledge Value dashboard showing per-issue PRDs, ADRs, lessons, and outcomes
Context compounds. Every issue auto-scaffolds a knowledge base — PRD, ADRs, lessons, retro outcomes. Future runs read prior context before touching code, so the system gets sharper with every PR.

Ready to see it in action?

Nightgauge is available to select clients. Get early access and experience the full pipeline.