Architecture / KynticAI Agentic Importance Framework
The local-first agentic layer over evidence-weighted importance.
KynticAI Agentic Importance Framework turns importance scoring into a private decision-support workflow: gather evidence, score what matters, identify missing data and contradictions, build compact model context, route to local loopback Ollama, an internal gateway, an OpenAI-compatible gateway, or an Anthropic-compatible gateway under deployment policy, then generate a safe answer, next action, or dossier.
Core layers
How KynticAI Agentic Importance Framework is structured
01
Evidence gatherer
Collect approved facts, summaries, source references, prior outcomes, and review notes from the local workflow boundary.
02
Importance scorer
Rank evidence by outcome relevance before the agent asks questions, routes work, or spends model tokens.
03
Question and contradiction layer
Identify missing variables, conflicting evidence, stale signals, and review needs before generation.
04
Context builder
Prepare compact model context with high-signal evidence, compressed secondary evidence, deferred low-signal material, provenance, and caveats.
05
Runtime router
Route to local loopback Ollama, an internal model gateway, an OpenAI-compatible gateway, or an Anthropic-compatible gateway under deployment policy.
06
Dossier reporter
Generate safe answers, next actions, forensic dossiers, missing-data notes, and provenance-aware reports.
Operating flow
The request path through the product
Gather
Gather evidence
Start from approved local evidence rather than asking an agent to scrape or improvise context.
Score
Score importance
Promote the facts that matter most and keep weaker evidence available without forcing it into every model call.
Ask
Ask for what is missing
Turn missing data and contradictions into better questions before the answer is generated.
Route
Route through the approved model path
Use loopback Ollama, internal gateways, OpenAI-compatible adapters, or Anthropic-compatible adapters under deployment policy.
Report
Report with provenance
Produce decision-support output with evidence references, caveats, missing-data notes, and review boundaries.
Monitor
Monitor the review state
Track caveats, review needs, and outcome feedback without turning product operations into the customer data plane.
Example signal path
A churn-risk agent produces a safe, provenance-aware dossier
Illustrative sample only. Deployment policy decides whether the model route is local Ollama, an internal gateway, an OpenAI-compatible adapter, or an Anthropic-compatible adapter.
01 / Source
Example fields
workflow = churn_risk_review
sources = CRM + support + usage
runtime = loopback_ollama
scope = customer_controlled
02 / Evidence
What KynticAI creates
importance_band = elevated
missing_data = renewal_owner
contradiction = usage_up + complaint_open
deferred = low_signal_notes
03 / Action
What the business does
generate next_action_brief
include missing_data_notes
retain provenance and caveats
route to human review
Operating model
How the product stays useful at enterprise scale
Weights
Protected scoring values
Outputs use reason categories, importance bands, and caveats while private scoring values stay inside the scoring boundary.
Tables
Private scoring tables stay protected
Generated dossiers receive context plans and explanation bands while proprietary scoring tables remain inside the protected runtime.
Runtime
Customer-controlled conversation data
Deployment policy decides whether work routes locally, through an internal gateway, or through an approved hosted-provider adapter.
Adapters
Approved model routes
Local Ollama, internal model gateways, OpenAI-compatible gateways, and Anthropic-compatible gateways are governed deployment options.
Integration points
Where it connects to the wider stack
Ollama
Loopback Ollama
Local model route for private agent proof and customer-controlled runtime paths.
Adapters
Approved provider adapters
Hosted providers such as OpenAI and Anthropic can be reached through approved OpenAI-compatible or Anthropic-compatible deployment adapters.
Workflow
Enterprise decision surfaces
Connect to internal review queues, sales/churn workflows, credit and hiring review, operations, supply chain, diligence, and support.
Dossiers
Safe dossier outputs
Return next actions, explanations, forensic dossiers, missing-data notes, and provenance-aware reports.