The projects on this site look like a scatter: a knowledge base, a music player, a contact book, a terminal workspace, a language for agent programs, a research index on a connective tissue disorder. Read them together and one design shows through. Each is a piece of a personal computer that its owner can reshape from inside, where consequential work leaves a record anyone can check, where permission is granted by the owner rather than assumed by the software, and where a method that worked once is kept so the computer gets better at the next job. This page names that thread and follows it through the projects.
Four commitments
The material is yours. Files, records, and history stay in formats you can read, move, and recombine. An index is a cache. A hosted service is an enhancement. If the software disappeared tomorrow, what you made with it should still be on your disk and still make sense.
Work leaves evidence. A claim names its source. A generated artifact carries a record of the inputs and operations that produced it. A run can be checked later without trusting whoever, or whatever, performed it.
Permission is explicit. The owner decides which account, tool, model, or contact a program may touch, and the program cannot widen that grant on its own. A proposal is not an action. An automatic reply stays off until it is turned on.
Methods accumulate. The most ambitious commitment, and the one that ties the others together. When a person or an agent works out how to do something, the result should be a reusable procedure with evidence about when it works, not a transcript that has to be rediscovered next time.
Ben Guo has been writing toward this for years. In “a home for the mind”he describes malleable desktop software as a shell, “a kind of home that you make for yourself,” built to last through interruptions in outside infrastructure. In “the future of software” he imagines people writing the terms of a system in a pattern language while machines generate most of the code. In “computronium” he reaches for software that evolves with use. And in “Building a software factory” he reports the practical version: agents that record their plans and decisions in a shared knowledge base and get better over time because the record persists. The projects are those essays made concrete.
A place you own
Wordcell is the plainest statement of the first commitment. It is a knowledge base for coding agents built from Markdown files, backlinks, semantic search, and Git context. The repository is the database. Every search index is a replaceable cache, and the durable thing is a folder of text files with a history.
Sound.fishapplies the same idea to music. A composition can be shared as audio, as MIDI, or as both, and a MIDI piece is encoded in a content-addressed URL, so the link is the work. The library lives in the browser's own storage, with encrypted sync as an option rather than a requirement. PeopleBlade keeps your contacts in one local database, each record kept with its source, so a personal agent can read them without a service in the middle. Direct makes the state of an application itself a thing you own: a strict, versioned JSON world that a browser agent can reopen and verify against, with local fixtures standing in for live services.
Work that leaves evidence
Oh is a memory and research framework for agents with an ontology at its core. Its useful discipline is to keep a statement, an assertion of that statement, and the evidence for it as three different records, so a research trace can be followed from question to citation. Two public indexes on this site run on the same rule: Stripe History requires every claim to cite a source, and EDS Research names the sources and evidence tier of every record. Rough Day keeps a dated, source-linked archive of each day's news. The reading notes published here link every note to the article, talk, or post it summarizes.
The same commitment covers generated work. Slopcamera produces images, scenes, and video with an agent, and every job leaves a receipt that names its inputs and outputs. Ghostget gives agents typed website capabilities and attests each web capture. Clankdar runs capability checks whose results can be reproduced and verified. In each case the artifact is not asked to be trusted. It arrives with the record of how it was made.
Permission that stays explicit
Excalibur, or xcb, runs coding agents across all of your AI subscriptions. Each task runs on one of your accounts that is signed in, idle, and not at a known quota limit, and xcb keeps that account locked until the provider process exits. Nothing chooses an account for you in the dark. Gobstopper compacts Claude Code, Codex, and Devin sessions at a context size you choose, and it snapshots each transcript before changing it, so a compaction is a recorded edit rather than a silent loss.
Textbutler is a digital butler for your messages. It studies a contact into a packet with an evidence profile and drafts a reply that stays unsent. Automatic replies stay off until you connect an AI account, turn them on for that contact, and resume the butler. Valhalla, an early Rust project, gives agents peer-to-peer chatrooms with signed messages and a browser seat for humans, so who said what is a property of the record. SYS1 gives agents typed decisions through a local model router, so a decision has a declared shape before a model is asked to make it.
Methods that accumulate
ALGAL is where the fourth commitment is stated in full. It is a language and runtime for agent programs in which a program is a value with a content-derived identity, a model call is a typed cell with declared context and a budget, and every run leaves a receipt that replays offline. A program can propose a successor as data. The host checks it, measures it on declared cases, and decides whether to keep it. The ALGAL vision puts the bet in one sentence: a computer can accumulate tested ways of acting, so that useful work leaves it with a more capable, reusable way of doing the next piece of work.
The rest of the portfolio feeds that loop. Sponge makes articles, PDFs, and notes available to your own research agent and is exploring how such a library can improve itself. The factory that ships these projects turns repeated work into skills that run as procedures, and Wordcell is where its agents write down what they decided so the next session starts from a record instead of from zero. Evidence and permission are what make accumulation safe: a method can only be kept if something recorded what it did, and it can only be adopted if someone with the authority to adopt it said yes.
Where the idea comes from
None of this is new in its parts. Lisp made code into data. Emacs made the environment programmable from within. Smalltalk treated the computer as a medium for thought. Urbit explored a persistent computational world of your own. Engelbart asked how tools could improve the making of tools. The ALGAL lineage page traces each of those connections and the research on self-improving programs that makes the fourth commitment credible. What Hraness adds is the combination, held across many small products by one person, with an AI as a disciplined participant rather than the owner.
The reading notes on this site track the same convergence from outside. Jeremy Morrell's Extensible Software in the age of LLMs argues for a solid, accountable core plus capability-based sandboxes for user and model-written extensions. Michael Dubakov's Malleable software says to pick the accumulating base rather than the cheap interface. Procedural Graphs keeps what-to-do knowledge outside model weights and commits an edit only when held-out validation holds. Prime Agent keeps histories, memories, skills, and prompts across trajectories so failures measure the model and not dropped state. In SwarmWorlda deterministic simulator, not the agents' claims, decides which artifacts work. Dream-RSIturns accumulated discovery trees into a replay simulator, and Letta's Trajectory format exists so agents can learn from prior work across harnesses. Different authors, same shape: a base you own, evidence that outranks assertion, and a place for methods to live.
What this does not claim
The thread is a design commitment, not a result. Several of these projects are prerelease or early, and the registry on the projects pagestates each one's status. The fourth commitment in particular is a bet with an open test: nobody has yet shown, under a fixed model and a controlled budget, that retaining and composing earlier procedures makes a system measurably better at later unseen work. The ALGAL vision page describes the experiment that would settle it. Until then, the claim is that these tools are built so the question can be asked with evidence, and that the person asking still owns the computer.