Cadora

The neutral conductor

One conductor for every coding agent — drive the top frontier models through a single workflow, on the subscriptions you already pay for, following the AI-DLC method, with proof you can ship.

open source · MITv0.9.0a capstone, not a platform

Cadora sits above the coding-agent CLIs and drives them through a workflow you declare — so the how of delivery stays yours. It is the conductor, not the agent: it doesn't implement the agent loop, and it isn't an in-IDE assistant. It's the layer that turns “an AI wrote it” into a run you can repeat, compare, cost, and prove.

what it gives you

Four things, one layer

01

Every frontier model

Claude Code, Codex, Kiro, and more through the same topology. A/B on identical work, or phase-split a run — design on one model, build on another. No lock-in.

02

Your own subscriptions

Subscription-funded by default — metered API is an explicit opt-in. Mix vendors and funding, with one cross-vendor cost ledger showing what each run spent.

03

A real method

Aligned with the AWS AI-DLC lifecycle — inception → construction → operations — a structured, repeatable workflow. Method-agnostic too: BMAD, or your own.

04

Proof you can ship

Deterministic gates decide “green,” not the agent. Every run is a signed, checksummed evidence pack — tamper-evident and attributable, verifiable after it leaves your machine.

how a run works

Declared, gated, recorded

You declare the steps; the gate — not the agent — decides whether each one is done.

visionspec
designinception
buildconstruction
gate ✓proven
signed packevidence

One topology, any model, on your own subscriptions — one cost ledger, one signed proof.

proof, not promise

Verify the pack

A run's evidence is checksummed and signed. Recompute the hashes and any change shows — an unsigned pack still verifies its integrity; a signed one is attributable too.

manifest.jsona3f1…9c
nodes/build/output.txt7be0…41
report.jsond29c…08
checksums.txt.sigssh-ed25519
— not run —
backends

Verified and experimental

verified · claude verified · codex verified · kiro experimental · glm experimental · antigravity

Every backend runs the same topology, gates, integrity scan, and archive — so results A/B-compare directly. cadora doctor prints each tier and version status.

honest status

Where this is

Cadora is a working reference implementation of the neutral-conductor idea — shipped, tested, backend-neutral. It is offered as the capstone of an exploration, not a finished commercial platform. The LLM-as-judge gate is a deliberate stub; the load-bearing parts — the gates, the integrity scan, the signed evidence, the cost ledger — are real. The vision stands on its own; the tool shows it holds.