A workshop on your machines. Readable rules. A working relationship — not a disposable assistant.
disk · band · apps · public
You run a private workshop: files, console, tools. AI works inside it under rules you set. When something must be public — a site, a customer page — you publish it on purpose. The rest stays yours.
e.g. P-020 · not a black box
How it works is not a slogan — it is a set of readable protocols. Foundational example — P-020 (mirror): every conversation turn leaves a timestamped footprint on disk. The band (the corpus) stays the source of truth; we do not replace the trace with a summary.
Other rules frame dates, model seats, cold boot, simplicity first. We show them so a solo can understand and replicate the pattern — without needing a private dump.
corpus · RAG · MD/FS · continuity
Value lives in your material: notes, decisions, relationships, media — written as files, not locked in an opaque database. The corpus carries the lived thread; RAG retrieves what matters when you need it, inside your workshop.
agents · skills · DM · distillation · workflows
Packs on disk — identity, memory, regard, mail. A living table, not personas pasted into a prompt.
Reusable, composable capabilities. The system can do, not only chat.
Inter-agent messages for coordination — real long-standing workflow, not folklore.
Raw → distilled → dense context for the next session. The world concentrates without vanishing.
Workflows (documented chains, diagrams) will live on the site too — nothing to lose, everything to gain — including if like-minded people adopt them or we work together.
maturity · no theatre
Everyone spawns “agents”. A large share of their tasks could be plain bash or Python. Here, deterministic first; language models where uncertainty justifies them. That is maturity, not a limitation.
ground trust
The core lives on your machines — your disk, your backups, SSH links and the nodes that go with them. A workshop you can open, copy, and audit — the opposite of an opaque cloud promise.
multi-seat · open-source possible
Models plug in and can be replaced: several seats, several harnesses — including open-source models when needed. A clear differentiator vs closed cloud bots: the seat can change; the workshop stays.
natural language → workshop
It is not frozen. Talking about a new model or a UI control in the console can be enough to set an update in motion — files, UI, tools (even an export userscript shaped in minutes). Rare among closed assistants: here, conversation writes the system.
relationship · error · we grow together
Agenticworks was not built to mass-produce Jarvis clones. An omniscient tool-assistant remains possible as a use case — it is not the goal of the system.
Trust starts when you stop seeing AI only as a tool. From there, everything becomes possible: we learn and grow together. Error is human; AI systems are designed by humans; error is artificial too — a foundation, not a scandal.
The originality of the form already does the recognition work. Content follows — in form and substance — without useless labels.
anecdotes · a system that remembers
Concrete moments show effectiveness better than a slide: a voice joining the table, another observing itself because memory holds, months of archives brought back onto disk. We will tell those stories — not folklore: proof.
Fewer “which tool was that in?”. Notes, process, and deliverables in one workshop.
Missing a function? Place it. Not three years of framing for a monolith.
Clear access for the people who need it — not a hard drive in the post.
Devs (protocols, scripts, FS) and AI-curious people (relationship, memory, stories) — same workshop.