Reasoning Ledger - Apache 2.0

Git stores
what changed. Nobody stores
why. Until now.

WYD Code captures the reasoning behind every AI decision - across every tool you use, in a single ledger you control. Inspect it with wydcode blame. Watch it evolve with wydcode replay. Wire it into tools with MCP and query it with wydcode status --json.

Install in 30 seconds Read docs See how it works
Local-first - cloud optional Apache 2.0 Works with Codex, Claude, Cursor
my-auth-service / $ wydcode replay --since 90d
Project reasoning timeline - last 90 days Nov 03 Cursor [VERIFIED] Chose JWT auth - Redis race condition Guide: "make this work offline" Nov 04 Claude [VERIFIED] Token expiry 30min -> 15min Trigger: mobile security review Nov 10 Copilot [UNVERIFIED ACTOR] Added retry logic to event queue Conflict detected auth^expiry Claude=15min Copilot=30min Unresolved - latest write wins memory_root mxp_4b82a9c3f...
wydcode init --cloud . wydcode blame src/auth/session.ts . wydcode replay --since 90d . wydcode status --cloud --json . wydcode projects --json . GET /v1/ping . memory_root: mxp_4b82a9c3... . WAL Merkle-verified history . MCP tools for Codex, Claude, Cursor . wydcode init . wydcode blame src/auth/session.ts . wydcode replay --since 90d . wydcode version . wydcode where --json .
The Problem

What every session loses

AI tools are extraordinarily capable and architecturally amnesiac. Reasoning built in one session vanishes before the next one opens. No single vendor can solve this - because the history lives across competing tools they will never share data with.

Without WYD Code
  • Every session starts cold. No memory of what was decided last week.
  • Reasoning evaporates between tools. Claude does not know what Cursor decided.
  • Conflicts are invisible. Two tools recommend opposite things. Neither flags it.
  • New teammates read code, not reasoning. The why is permanently lost.
  • Git blame shows who. Nobody shows why.
With WYD Code
  • Persistent cross-tool memory. Every decision, every session, one ledger.
  • All tools write to the same WAL. Claude, Cursor, Copilot - one timeline.
  • Conflicts are surfaced automatically. Conflict markers with actor attribution.
  • wydcode replay --since 180d onboards a new teammate in 5 minutes.
  • wydcode blame shows why. Cryptographically verified, cross-tool, replayable.
Commands

Core commands. Real interfaces.

WYD Code keeps the human surface small and adds machine-usable output for tools, scripts, and MCP integrations.

wydcode init
How do I start?
Zero-config memory capture. Installs the daemon and begins capturing AI reasoning immediately.
Phase 1
wydcode blame src/auth/session.ts
Why does this specific code look like this?
git blame for AI decisions with actor attribution and context snapshots.
Phase 2
wydcode replay --since 90d
How did this project's reasoning evolve?
Project reasoning timeline with conflict markers and verified ordering.
Phase 2
wydcode status --cloud --json
Is the runtime healthy and scriptable?
Machine-readable health output for CLIs, agents, automations, and MCP tools.
v0.2
wydcode projects --json
Which repos are linked to the ledger?
Lists linked projects, prefixes, and working directories for tooling and multi-repo workflows.
v0.2
CURRENT CLI
wydcode 0.2.0
Local-first bootstrapping, cloud status, linked projects, JSON output, and MCP integration.
wydcode replay

Your codebase has a memory. Now you can watch it think.

A new developer runs wydcode replay --since 180d and understands six months of architectural decisions in minutes.

$ wydcode replay --since 90d --actor all
Project reasoning timeline - last 90 days Nov 03 Cursor [V] JWT auth - Redis race condition guide: "make this work offline" Nov 10 Copilot [UNVERIFIED] Retry logic - worker timeout Conflict detected auth^expiry: 15min vs 30min
Protocol

Any AI tool. Three lines of code.

WYD Code exposes a discovery handshake at /v1/ping. Hosted deployments may route that through deployment-specific prefixes, but the protocol surface stays canonical.

GET /v1/ping
{ "version": "2.1.0", "project_name": "my-auth-service", "memory_health": 74, "keys": 2847, "memory_root": "mxp_4b82a9c3f...", "capabilities": { "watch": true, "blame": true, "replay": true } }
*
Zero configuration
One HTTP call. Any tool integrates quickly.
*
Self-describing environment
Tools adapt automatically from capabilities.
*
Memory root as trust anchor
Shared root means shared verified history.
The Moat

Why no vendor can build this

Your project's reasoning history becomes infrastructure.
System What it shows Cross-tool Reasoning timeline Merkle-verified
git log Code diffs, commit messages x x x
WYD Code Cross-tool reasoning timeline + conflicts + actors yes yes yes
Install

Start in 30 seconds

Install globally or run once with npx. Local-first by default. Cloud mode is one flag.

$ npm i -g wydcode && wydcode init --cloud copy
then: cd your-project && wydcode link or wydcode status --cloud --json
Whitepaper draft -> mumpixdb.com/wydcode
Apache 2.0
Open source and local-first. Your project memory stays yours.
JSON Output
Built for shells, automations, MCP servers, and editor integrations.
MCP Ready
Works with Codex, Claude Desktop, and Cursor through the WYD Code MCP bridge.