A Bitspace Applied Intelligence project

Do Your Focused Job, AI.

DYFJ is a sovereign AI stack: modular, vendor-loose, local-first by default, with cost visibility built in.

View the repo Early and active. Not a hosted SaaS.

What it is

An AI stack without the handcuffs.

DYFJ is a first-class AI stack: local-first by default, modular behind stable contracts, and designed so model choice, runtime choice, memory, permissions, and cost are visible parts of the system instead of convenient afterthoughts.

It is early, active, and deliberately not a hosted SaaS. The point is agency: durable context, swappable components, cost-aware choices, and tools that help the operator know what happened.

dyfj session log illustrative

$ dyfj run "summarize today's substrate work"
→ model: local/qwen3-32b   runtime: ollama   est. cost: $0.00
✓ done in 4.2s — 1,812 tok in / 246 tok out · session cost $0.00
$ dyfj log --last 1
2026-09-10T19:44:02Z  run.complete  model=local/qwen3-32b  tok=2058  cost_usd=0.00
  memory: 3 facts derived from event log · log is ground truth

Operating stances

Local-first by default

Inference, memory, and tooling should start close to the operator. Hosted or paid calls should be visible, attributable, and controllable.

Vendor-loose, not vague

Components can be replaced behind stable contracts. Strong defaults matter; lock-in does not get to be the architecture.

Cost is visible

Token spend, model selection, and budget posture belong in the working surface before and during the work.

The log is ground truth

Memory is a derived view. The message and event log are the durable audit trail for what the system actually did.

Source

The repo is the current source of truth.

The DYFJ repo carries the current substrate work: prototype code, schema decisions, and the operating posture for the project. This site exists to point you there.

View the repo →