Skip to content

M2a — Live standalone chat #2

Description

@mdheller

Goal

Make Noetica demonstrably work end-to-end in standalone mode without requiring SourceOS or Neuronpedia credentials.

DONE when

  • A message sent to Claude Sonnet or an OpenAI model returns a real response from the provider.
  • Governance trail shows model_routed, provider, and latency_ms from the actual provider call.
  • Typing indicator appears during inflight request.
  • Steering panel handles all model capability tiers cleanly:
    • full: steering controls visible but missing Neuronpedia key returns status: not_configured.
    • local: clearly marked as requiring SourceOS / Agent Machine local inference.
    • none: blackbox provider path shows provenance/tamper-evidence posture, not SAE steering.
  • No NEURONPEDIA_API_KEY is required for M2a completion.
  • CI green on the merge commit.

NOT in M2a

  • Real steering diff. That is M2b.
  • SourceOS routing. That is M3.
  • Memory-mesh integration. That is M3.
  • Google / Meta / Mistral adapters. They may be added later but do not block M2a.

Boundary rule

M2a must not represent prompt steering, system prompts, or few-shot examples as SAE steering. Blackbox providers are provenance-only from a mechanistic-steering perspective.

Acceptance evidence expected

  • PR link.
  • CI run link / status.
  • Manual smoke notes showing at least one successful Anthropic or OpenAI standalone response with governance trail fields populated.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions