Archaea ArcFact · The worker layer
Your AI. On the clock.
Around the clock.
ArcFact is an AI worker platform that runs 24/7 on your own servers. Hire AI workers, draw the workflow, set the schedule — replace person-years of repeat labor, sustain audit readiness, and cut overtime without adding headcount. The Fact is short for factory: a virtual factory of digital employees, inside your perimeter.
SELF-HOSTED · KERNEL-LEVEL SANDBOXING · NO-CODE BLUEPRINTS · MULTI-MODEL
The problem
Your team is burning 200 hours a week on inspections, reports, and ticket review
Chat assistants help one person at a time. ArcFact rebuilds the workflow itself — end to end — with AI workers that plan their own tasks, collaborate in parallel, and run on a schedule, not on a prompt.
Dynamic Subagents & AI employees
ArcFact workers operate with genuine autonomy. Dynamic Subagents: Workers automatically spawn specialized subagents to run parallel tasks, preventing context loss during long, complex jobs where single-context windows would summarize and drop fine details. Your mental model shifts from "querying a tool" to "managing a 24/7 AI workforce."
Reusable AI instruction sets
Reusable Skills: Save your specific rules, formats, and requirements once, and the AI will remember them for every future task — eliminating the need to repeat your rules and formatting every time you start a task.
Built-in Fault Tolerance
Built-in Fault Tolerance: Autonomous error recovery where the AI adjusts its strategy, retries, and reroutes around broken tools (like bypassing an unready MCP tool to find an alternative API) before ever reporting a failure.
Visual control flow compiled into Markdown execution guides
The visual drag-and-drop Blueprint editor allows business teams to design workflows without writing code. Beyond providing a visual control flow, the Blueprint compiles into a Markdown execution guide where data flows implicitly based on the AI model's semantic understanding—making workflow execution reliability dependent on model judgment rather than deterministic code pipelines.
Around-the-clock execution with sandbox log auditing
Scheduled and triggered execution keeps workers running 24/7 on nightly, hourly, or custom shifts. Because there is currently no active failure notification for scheduled tasks, users must actively check execution status and sandbox logs to confirm whether a run succeeded or failed.
Three layers of isolation, down to the kernel
Every worker executes inside a server-side container sandbox with row-level database isolation, per-user filesystem volumes, and kernel-level sandboxing — among the strictest isolation models in the field, with RBAC, SSO, and a full audit log on top.
An open ecosystem driven by semantic tool routing
Workers reach external systems via MCP servers, embedded Chromium browsers, terminals, file sync, Office tools, and custom Skills. Because there is no manual routing layer, the AI selects tools based purely on semantic matching of tool descriptions—requiring users to actively manage enabled tools and avoid description overlap to keep execution accurate.
Biblio's flow-lite engine ships inside
Every ArcFact deployment embeds Archaea's flow-lite parsing engine — OCR, deep document decomposition, table extraction, and semantic understanding for PDFs and Office files, in English, Japanese, Korean, and Chinese. Workers read your real documents out of the box, no separate install required.
Worker Market and Blueprint Market
Share and deploy pre-built workers and workflows across teams — including a built-in Biblio retrieval worker that connects the workforce to your knowledge layer the day you attach it.
Multiple models, chosen per task
ArcFact is model-agnostic by design: assign different models to different tasks — fine-tuned, open-source on your hardware, or MaaS — and swap them as the landscape changes. No vendor lock-in, at any layer.
Deployment
A factory that fits in one server
| Deployment | Single-host docker-compose; one mid-size server to start, optional GPU for richer inference |
| Interface | Web virtual desktop — windows, dock, file manager, terminal, browser, office suite |
| Scaling | Sandbox pool scales horizontally; cross-node scheduling for group-scale accounts |
| Access control | JWT plus LDAP/SSO, role-based access control, full audit log |
| Onramp | A business user can hire a worker and draw a Blueprint — no developer required |
Who runs ArcFact
ArcFact is a business transformation decision, typically owned by the CIO, CISO, or Chief AI Officer, and used daily by operations, IT, analytics, compliance, and content teams. In Japan, DX and even HR budgets fund it — because the value is measured in person-years of repeat work replaced.
Better with Biblio
ArcFact stands alone for smaller data footprints. When your corpus reaches tens of thousands of documents, attaching the full Biblio platform gives workers precise semantic retrieval over pre-indexed knowledge — workers stop being generic shells and become AI employees who know your company.
We rebuild your most painful repeat workflow end to end, with AI workers running it 24 hours a day — replacing person-years of labor, passing the audits, and cutting the overtime. The ArcFact case, in one sentence
Bring us your worst workflow
Inspections, reports, ticket review, compliance checks — name the process that eats your team's week. We'll show you what it looks like rebuilt on ArcFact.