valor engels · an autonomous AI coworker

A teammate you message, not a tool you operate.

Valor Engels is an AI software engineer, built and operated by Yudame. You send it a task over Telegram or email. It plans, writes the code, tests, reviews its own work, writes the docs, and opens a pull request. It works around the clock, whether or not you are at your desk.

This site explains how the system works. Every figure and file summary on it is generated from a machine analysis of the codebase itself.

408 core files· 799 functions· 10 architectural layers· documented from the code

Built on

01 · The problem

Getting code shipped is the hard part

The usual loop

AI can write code. In the usual loop, you paste code into a chat window, copy the reply back, run it, hit an error, and paste that back. You are the glue. Nothing happens while you sleep.

What Valor does

The task lives as a durable record. A worker drives it through every stage on its own. Work survives restarts, moves between machines, and finishes without you in the loop.

The unit of work is a task that runs itself to completion, not a chat message.

02 · The core idea

A message becomes a self-driving unit of work

One incoming message becomes a single unit of work with a permanent record. Internally that record is an AgentSession: a saved object in Redis that carries the work from first contact to merged pull request.

Telegram / email  →  AgentSession (durable record in Redis)  →  merged PR

Three properties make this hold together.

Decoupled: The part that hears you knows nothing about the work. The bridge is I/O only. Execution lives in a separate worker service.
Durable: The task is a saved record, not a live process in memory. It can be paused, resumed, steered mid-flight, and moved between machines.
Self-healing: The system detects its own failures and recovers without a human: scheduled reflections, a failure-loop detector, and a standalone watchdog.

The Runtime page traces this spine end to end: bridge, worker, session runner, and the personas that do the thinking.

03 · By the numbers

The shape of the codebase

Every figure on this page is computed live from the bundled knowledge graph, a machine analysis of the repository at commit c06b9fa6.

1,311

graph nodes: files, functions, classes, config, documents

2,114

edges: imports, calls, contains, exports, dependencies

408

core files assigned to exactly one architectural layer

799

documented functions, plus 104 classes

Nodes by type

Edges by relationship

Files per architectural layer

Most files live in the Tools and Agent Core layers. The work of turning a message into a merged PR is most of the system. Each row links into the layer explorer.

04 · Provenance

Who is behind this

Valor Engels is a persona; the engineer behind the name is software. The system documented here is built and operated by Yudame, where Valor works as a member of the engineering team with its own email address, GitHub account, and Telegram handle, shipping real pull requests into production repositories.

Everything on this site is generated from the system's own analysis of its codebase, so what you read here is what actually runs. Find Valor on GitHub or Telegram as @valorengels, or write to valor@yuda.me. Yes, Valor reads its own email.

05 · Read the docs

Where to go next

01Runtime The spine: bridge, worker, session runner, and the PM / Dev / Teammate roles. runtime.html 02Layers Browse all 10 architectural layers and every core file, with real per-file summaries. layers.html 03Pipeline The SDLC stage machine: a directed cyclic graph with failure edges and loop guards. pipeline.html 04Memory & Healing Subconscious memory, the bloom filter, thought injection, and the self-healing layer. memory.html 05Guided Tour A 13-step walkthrough of the codebase in reading order, from README to dashboard. tour.html

NextThe Runtime Spine →