Foundry OS & Assurance Plane · In development

Training is a branch, not an automatic step.

Foundry OS is designed to take an organization from an initial requirement to a controlled deployment—connecting measurable acceptance criteria, model/license/compute selection, data preparation, conditional method choice, evaluation gates, human approval, deployment, and governed updates.

Requirements-to-production lifecycle

Interactive Lifecycle Map · Click any stage to inspect technical detailTraining is a conditional branch

▲ CONTINUOUS ASSURANCE PLANE (SPANS ALL 7 STAGES)Data Rights · Provenance · Security · Evaluation · Approval Gates · Release & Rollback

STAGE 04 TECHNICAL DETAIL · Choose least-complex method

Branch only as far as the problem requires—training is not automatic.

Compares (A) Prompt + Retrieval, (B) Adapt Existing Model, and (C) New Architecture against the same baseline.

Branch A · Prompt + Retrieval

Selected when existing model weights plus structured context meet acceptance criteria.

Branch B · Adapt Existing Model

Selected when domain terminology, schema strictness, or cost/latency bounds require fine-tuning or post-training.

Branch C · New Architecture

Selected when Research OS investigation is justified by structural or efficiency constraints.

Interactive console walkthrough

ARCENCE CONSOLE WALKTHROUGHIllustrative planned workflow

Natural brief ⇄ editable structured constraints → proposed plan → human approval → governed execution

Load example environment:

Detail level:

1. What are you trying to build?NATURAL LANGUAGE FIRST

Accepts plain-English goals, pasted notes, or protocol constraints. YAML is generated automatically—never required upfront.

2. Clarify the decisions that change the pathSTRUCTURED CONTROLS

Data Boundary

Evaluation Priority

Operating Environment

Level of Control

3. Compare viable paths (least-complex first)CLICK TO SELECT PATH

PROPOSED MODEL-ENGINEERING PLAN

4. Review Plan & Approval Gates

Simulated · Ready for approval

SELECTED WORKFLOW PATH

Path A: Prompt + Source-Linked Retrieval (No Training Required Yet)

Evaluates existing compatible models with retrieval first so your team avoids unnecessary training complexity.

DATA & RIGHTS BOUNDARY

Tenant-isolated Customer Learning Vault. Data remains customer-controlled and is never pooled or resold.

EVALUATION & HUMAN APPROVAL GATES

Primary acceptance check: Source citation accuracy & audit trail completeness
Target operating boundary: CUSTOMER VPC (subject to compatible licenses and dependencies)
Requires explicit human sign-off before promotion or release packaging.

GENERATED HANDOFF ARTIFACT (CLICK TO EDIT)

# Arcence Project Spec (Generated · Editable)
project_brief: "I need an internal assistant that helps our operations and legal teams analyze commercial contracts, cites the exact policy or clause for every flag, and stays inside our customer VPC."
data_boundary: private_vault
evaluation_priority: citation_audit
deployment_environment: customer_vpc
control_mode: guided
recommended_path: prompt_rag
assurance_gates:
  - baseline_comparison_required
  - license_and_provenance_check
  - human_approval_before_release

Deterministic local simulation · No external model or job executedFirst bounded wedge: Arcence Model Preflight

Deployment continuum

Deployment Continuum · Customer Control Increases Left to Right ───►Click any environment boundary

SELECTED BOUNDARY · Customer VPC

Deploy inside your organization’s dedicated AWS, GCP, or Azure Virtual Private Cloud with private networking and customer-managed keys.

COMPATIBILITY REQUIREMENTS

Compatible cloud provider IAM, container runtime, and license terms.

Note: Availability depends on compatible licenses, runtimes, and local dependencies.