Long-form writers who want a focused daily-driver editor built around their hands.
PerfectStar 2K
A keyboard-first writing environment that combines WordStar navigation, Reveal Codes, persistent sessions, spellcheck, and manuscript-ready RTF export.
AI software factory
Software that ships. A factory that compounds.
sXs builds products, reusable capabilities, and selected client software through a production system that combines explicit specification with model-driven execution. Verification and accountable judgment decide what ships.
SpecifyGenerateVerifyShipCompound
Latest shipment
Buildingv0.5.2 ·
Archipelago
A federated text-world engine for ergodic literature: sovereign realms authored in Markdown, played over SSH or the browser, whose events become part of a permanent world record.
What changed
The current release runs the local realm end to end, with federation foundations and sea travel in place.Products
The software is the output.
No two of these products share a language, a runtime, or an audience. AI may be inside the product—or only inside the production system that made it.
Readers who want a text world to inhabit, and authors and operators who want to run a sovereign one.
Archipelago
A federated text-world engine for ergodic literature: sovereign realms authored in Markdown, played over SSH or the browser, whose events become part of a permanent world record.
Researchers and scholars who need to know where their sources disagree and which conclusions rest on a single paper.
Datalinks
An agent wiki that turns a Zotero library and agent session logs into a maintained research vault where every assertion traces to a source it can verify.
Factory capabilities
Every build should leave the factory stronger.
Reusable agent knowledge, decision systems, and domain workflows turn repeated learning into infrastructure for the next product.
Engineering teams that need reusable agent knowledge without maintaining a separate source for every harness.
Kanon
A canonical knowledge-artifact system that validates and compiles skills, rules, workflows, agents, and prompts for multiple AI coding harnesses.
Teams that need consequential technical decisions to remain visible to both people and agents.
ADR Power
A portable agent capability for creating, reviewing, cross-referencing, and maintaining architecture decision records inside a codebase.
Writers and publishing professionals who need repeatable, inspectable AI-assisted craft workflows.
Byron Powers
A library of literary and publishing workflows for novelists, technical authors, agents, publicists, and proofreaders.
Production model
The model may propose. Evidence decides.
symbolic × subsymbolic is an operating model: explicit structure where the product must keep a promise, learned synthesis where judgment creates leverage.
- 01
Specify
What must be true?
Turn product intent into constraints, interfaces, acceptance criteria, and consequential decisions.
- 02
Generate
Where does synthesis help?
Use models and specialized agents for exploration, implementation, transformation, and documentation.
- 03
Verify
What evidence decides?
Build, test, inspect, and review generated work against the intended product behavior.
- 04
Ship
Can someone use and own it?
Release operable software with documentation, deployment, and a clear maintenance path.
- 05
Compound
What should the next build inherit?
Convert repeated knowledge into capabilities, templates, components, and decision patterns.
Factory evidence
Artifacts, not productivity theater.
Releases, architecture decisions, tests, and inspectable capability specifications make the production system legible without inventing a velocity multiplier.
Release
Kanon v0.5.0
A versioned release of the factory’s canonical knowledge-artifact compiler and catalog.
↗Decision recordKanon architecture ledger
Public ADRs record decisions about adapters, validation, security review, distribution, and harness support.
↗VerificationPerfectStar 2K test surface
The product repository documents unit coverage for buffer behavior, undo, wrapping, Markdown, and RTF export.
↗Operating artifactADR Power specification
The capability makes its workflow, invariants, decision states, and edge cases inspectable.
↗BoundaryArchipelago status boundary
The project states which subsystems run today and which remain unfinished, so its documentation cannot be read as a shipping claim.
↗VerificationDatalinks corpus invariants
Six invariants are stated once and enforced by a validator, so a generated wiki cannot cite its own reasoning as a source.
↗Factory Notes
Publish when the work produces evidence.
Release-driven notes on builds, capabilities, boundaries, verification, and the parts of the factory that genuinely compound.
01 / FN
The software ships. The factory compounds.
What sXs means by an AI software factory—and why the output matters more than a claim about speed.
Read Factory Note02 / FN
PerfectStar 2K: what a non-AI product proves about an AI software factory
A keyboard-first writing environment shows why AI-native production should be judged by the software, not by whether the product contains a model.
Read Factory Note03 / FN
Kanon: turning reusable agent context into factory infrastructure
Kanon gives the factory a canonical source for skills, rules, workflows, agents, and prompts across multiple AI coding harnesses.
Read Factory NoteWork with the factory
Bring an outcome the factory can own.
Selected builds and partnerships come first. Advisory is available when a clear decision, not more software, is the useful output.
Commission a Build
A bounded product or system with explicit ownership, acceptance criteria, and a release target.
Adapt a Capability
Apply or extend existing factory infrastructure inside a new product or environment.
Factory Partnership
A longer collaboration for products that benefit from repeated releases and shared learning.
Decision Review
A standalone product or architecture decision with a written artifact. Advisory, when a build is not yet the right move.
Direct contact
Start with the product and the constraint.
A useful first note names the intended user, current state, principal constraint, and desired release horizon.
[email protected]