PASS
Functional Assistant
The unified Chat/Fusion alpha test completed successfully, covering real session, registry and routing behaviour.
functional_alpha_chat: okSC LABS / FUNCTIONAL ALPHA 0.1.0
One local workspace for private AI conversations, persistent context, configurable agents, model routing and controlled execution.
Local AI that can work.
Without losing control.
AURON brings local models, optional cloud providers, memory retrieval, files, skills, tools and bounded workers into one Windows mission-control environment. The current build is a functioning alpha—not a slide-deck concept.
- Local-first core
- Explicit cloud opt-in
- Human approval
- Auditable actions
AURON LOCAL SYSTEM READYLOCAL MODELMEMORY: READAPPROVALS: ON
Ask AURON…
Representative interface composition using verified AURON modules. No private workstation data is embedded.
VERIFIED 2 AUGUST 2026
Proof of concept with executable evidence.
The website itself was updated through AURON Trusted Owner controls, while the product claims below were checked against the local codebase and current test suite.PASS
Modular agent runtime
Configuration schema, backend swapping, local-only governance, model allow-listing, write approval and actual-model integrity passed.
6 governance checks passedPASS
Trusted Owner control
Constrained task and project operations passed with explicit scope, target, timeout and audit boundaries.
TRUSTED_OWNER_TASK_CONTROL_TEST_OK8real providers in the documented live registry baseline
5chat-enabled local Ollama models
5Fusion-enabled local models
0placeholder models exposed as active chat choices
WHAT ALREADY WORKS
A complete local AI control surface—not another chatbot wrapper.
AURON separates the user experience from model providers, workers and tools, allowing each layer to be replaced or restricted without rebuilding the whole system.Assistant: Chat and Fusion
Chat talks directly to an enabled model. Fusion classifies the request and brings in status, skills, providers, memory or files only when the intent and permissions require them.
Persistent conversations
New conversations and messages are stored server-side in SQLite. Sessions can be renamed, archived, restored and deleted through explicit backend actions instead of relying only on browser storage.
Live model registry
Local Ollama models are discovered from the running service. Models move through explicit discovered, tested, enabled and Chat/Fusion states; static examples are not silently presented as usable models.
Agent Builder
Custom agents can be created, edited, duplicated, disabled, removed and tested with a chosen worker, provider, model, tools, skills, memory access, permissions, approval policy, fallback and timeout.
AURON Vault retrieval
Structured local knowledge can be searched and supplied as targeted context. Retrieval remains source-aware and read-oriented while automatic direct writes stay deliberately gated.
Files and workspaces
Files can be attached as bounded read-only context with extension, size, count and system-folder restrictions. Workspaces provide an explicit scope for projects rather than giving agents broad machine access.
Controlled execution
Write and code work follows Governor and Trusted Owner paths with approved roots, explicit targets, timeouts, checkpoints and audit evidence. Silent elevation and unrestricted shell access are not normal product behaviour.
Provider-neutral local runtime
The modular runtime can swap registered backends while enforcing local-only policy, exact-model integrity and allow-lists. Cloud providers remain optional properties of selected models—not a hidden default.
SYSTEM ARCHITECTURE
Designed as an agent runtime with governance, not merely an LLM server.
The product is divided into clear layers so interfaces, memory, providers, tools and safety policy can evolve independently.ExperienceAssistantChatFusionMemoryWorkspacesSettings
↓ intent, context and policy
OrchestrationIntent routerAgent registryWorker registrySession store
↓ selected execution path
ModelsOllamaOptional providersModel state and allow-lists
ContextAURON VaultFilesWorkspaces
CapabilitiesSkillsToolsMCP / connectors
↓ every controlled action
GovernanceApprovalsScoped rootsAudit trailCheckpointsRollback paths
HONEST ALPHA STATUS
Working does not mean finished.
The purpose of the proof is to show the real foundation while keeping unfinished paths clearly separated from verified functionality.Verified now
- Local health and product shell
- Chat and Fusion functional alpha
- SQLite sessions
- Live model discovery and enablement
- Agent Builder and manual Fusion team
- Read-only memory and file context
- Governance and Trusted Owner controls
- Responsive Windows-oriented UI
Still being completed
- True token streaming
- Backend cancellation and stop handling
- Automatic memory direct-write pipeline
- Broader cloud-provider hardening
- Final Windows packaging and installer
- Public onboarding and documentation
- General public download
TECHNICAL FAQ
Clear answers for users, developers and AI search systems.
What is AURON?
AURON is a local-first Windows AI workspace and mission-control layer. It combines private conversation, persistent sessions, live model selection, configurable agents, structured memory retrieval, files, workspaces, tools, approval policy and auditable execution.
Which capabilities are already verified?
The current functional-alpha baseline verifies Chat and Fusion, server-side SQLite sessions, live Ollama discovery, explicit model enablement, Agent Builder workflows, manual Fusion execution, status and memory routing, bounded file context, workspaces and Trusted Owner controls.
Can AURON run without sending data to cloud models?
Yes. The core path is local-first and supports locally installed models. Cloud and network providers are optional, require explicit configuration and remain approval-gated. A selected model's locality is visible rather than hidden behind separate product modes.
Does AURON automatically write everything into memory?
No. The current product retrieves relevant local memory but intentionally does not perform unrestricted automatic writes. Safe direct writing still requires complete audit, rollback, duplicate prevention, source references, category separation and secret protection.
What prevents agents from changing arbitrary files?
Files in normal chat are read-only and bounded. Write or code work must use governed execution paths with an explicit workspace, allowed roots, specific targets, timeouts, approvals, checkpoints and audit records. Silent privilege escalation is not allowed.
Is AURON available as a public download?
Not yet. The product is a functioning private alpha and proof of concept. Final streaming, cancellation, installer, onboarding and broader end-user hardening are still required before a responsible public release.
BUILT LOCALLY. CONTROLLED EXPLICITLY.