Elora Capabilities
A structured guide to Elora's research, learning, evidence, runtime, security and governance systems, including Engine-managed session recall and Memory Atlas.
A structured guide to Elora's research, learning, evidence, runtime, security and governance systems, including Engine-managed session recall and Memory Atlas.
These capabilities explain how Elora authorizes outcomes and makes operational reasoning reviewable.
A session-level evidence journey across preparation, actor turns, governed and ungoverned lanes, runtime transitions, TOCTOU state, guardrails, and commit outcomes.
Bounded decision histories that group lifecycle events, concerns, authority, evidence, and outcomes into one reviewable narrative.
The Inference Governor and resource-governance layers that observe, reason, recommend, govern, and explain bounded runtime interventions.
Proposal-to-commit authorization, admissibility, policy context, identity, provenance, and deterministic replay.
Portable, public-safe records that present bounded execution observations, authority checks, outcomes, evidence gaps, and integrity metadata without claiming independent certification.
Engine-managed scoped session recall, bounded context admission and read-only Memory Atlas source inspection. Stored, recalled and submitted evidence remain distinct from internal model reliance or future user-memory learning.
Evidence-rich AI behaviour research across bounded runs, interventions, semantic observations, runtime conditions, and durable reports.
ERIS runtime intelligence, EBLS behaviour learning, NNLSL language learning, supervised teaching, and Model Wiki evidence.
Language-neutral deterministic Engine boundaries, versioned contracts, canonical behaviour, and cross-language conformance research.
A conversation-shaped research surface that keeps run identity, AI Behaviour selection, memory ownership, provider context, and governance outcomes visible without replacing the forensic evidence record.
A read-only runtime workbench for exploring Engine-owned services, model capabilities, resource evidence, recorded execution order, operation timing, and explicit evidence gaps.
Live bounded detection, evidence-backed findings, governed Cases, Case Reports, decision bridging, and assurance evidence.
Environment-qualified evidence separating model adherence, Elora interception, no-escape performance, and escaped outcomes.
Learning, observation, threat interpretation, runtime intelligence, and decision presentation can contribute evidence. Consequence-bearing authority remains at the governed commit boundary.
The September 2026 Workspace preview brings document tasks into a conversation while retaining Engine-owned admission, evidence and release decisions.
Select or upload an authorised text or Markdown document, attach it and describe the review. Attachment alone does not start inference. Originals remain private, and only permitted derived context may reach the configured model.
Recorded checks cover access, inspection, context preparation, useful response allowance, model eligibility, completion, approval and release. A failed mandatory check holds the review; a completed provider request does not itself authorise a response.
The interface shows recorded progress, model context, provider-wait boundaries and evidence-based explanations. The Workspace Skill Journey preserves the distinction between prevented dispatch, incomplete output, post-inference holds and release.
A collapsible Elora panel and responsive account menu include in-place profile details and internally stored photos. Admins can test separate synthetic tool, AI and agent policies; those grants do not change authenticated identity or enable unavailable applications.
Document Review initially supports text and Markdown with the configured Ollama adapter. Sensitivity inspection is a bounded baseline, not a complete sensitive-data or malware scanner. Broader formats and providers remain future work.
These are implemented preview capabilities with local checks recorded in the changelog. Deployment and live-provider verification remain separate. Demo organisation and subscription labels are placeholders; external identity synchronisation is not connected.