Source intake
Brings project material into structured, provenance-aware records.
The Fort · private rebuild
Durable project memory — not an untracked chat transcript.
An owner-controlled AI development environment for building real applications with durable project memory, scoped tool access, source-backed context, and reviewable execution.
Private rebuild Early access is discussion-based, not a live self-serve deployment.
Five commitments
Decisions, proof, and context persist as project knowledge — not a chat you lose.
Agents work from organized, provenance-aware sources, not an ambiguous prompt dump.
A bounded, permissioned gateway replaces open-ended automation.
Work is proposed with evidence and gated behind an explicit operator decision.
Sensitive values stay out of AI-readable artifacts; the default posture is deny.
The problem
When memory is a chat window and permissions are broad, work becomes hard to trust and harder to audit. The Fort changes the posture, not just the prompt.
The approach
The Fort gives agents the right context before they act, binds useful work to scoped boundaries, and keeps the operator in control of what changes, runs, and ships.
It is not framed as open-ended automation. The direction is a private development layer where knowledge, retrieval, tool access, validation, governance, and review reinforce each other.
Memory carries forward into the next run.
Developer workflow
Organize project material into reusable context without promoting old folders wholesale.
Give the agent focused context instead of an ambiguous prompt dump.
Keep changes bounded to the intended module, file set, or investigation.
Architecture snapshot
The public view is intentionally high level: enough for technical evaluation, not enough to expose internal mechanics.
Brings project material into structured, provenance-aware records.
Shared contracts for artifacts, source references, retrieval, and refusals.
Supplies task-specific context without an open-ended raw query surface.
Checks that work has evidence, boundaries, verification, and handoff.
Presents review, status, and decisions to the human operator.
Protected paths, owner approval, staged arming, and signing.
Safety & governance
Owner-held sensitive values stay out of public artifacts, access is scoped, and proposed work is not treated as accepted just because an agent produced it.
Current status
The knowledge foundation has been promoted as the canonical first-module baseline. Additional modules are proposed integration baselines under reconciliation. Private deployment, live tool exposure, production knowledge storage, and app dogfooding remain gated on later decisions.
Next step
Early conversations are best with technical founders, builders, and operators who already feel the limits of ordinary AI coding tools.