Inject

Any data item

Connectors or approved one-off imports load email addresses, cookies, events, CRM rows, products, support, billing, documents, and outcomes.

Store

Attribution paths

Scout stores each item's relationship set, vector, and ordered attribution path in customer-owned PostgreSQL/pgvector. Fortress moves high-load memory into Rust/LanceDB, not KynticAI product operations.

Compare

Rust + LanceDB

Enterprise contains the proprietary Rust relationship, weighting, traversal, and LanceDB path for fast similarity analysis across millions of relationship sets.

Output

Top-example JSON

The engine creates JSON with the best examples, importance bands, confidence, caveats, and ranked task options for the question being asked.

Explain

LLM task brief

Fortress sends JSON to your approved model boundary. The proposed Elite path shows the executive walkthrough from safe discovery to scoped pilot and outcome review.

Product / KynticAI Agentic Importance Framework

KynticAI Agentic Importance Framework

Agentic workflows powered by evidence-weighted importance scoring.

Gather evidence, score what matters, ask better questions, compress context, route work, and generate safe dossiers.

Private decision-support agentsSales and churn teamsCredit and hiring teamsOperations and supply chainDiligence and support teamsLocal-first product teams

Agent run / Framework

The Agentic sale is seeing a private agent pause, score, ask, and then act.

The buyer should feel the difference immediately: this is not an agent that grabs more context and writes more text. It ranks evidence, asks for the missing variable, and gives the team a reviewable dossier.

1. Agent receives the job

request = Build a renewal-risk dossier for Northstar Account

sources = CRM, support, usage summary, renewal calendar

target = next action for customer success

2. Framework scores the evidence

high_signal = open severity-two ticket

high_signal = usage down over the current review window

medium_signal = sponsor asked for security proof

deferred = old campaign engagement

3. Agent asks before it guesses

missing = latest engineering update

contradiction = usage down but executive sponsor still active

question = confirm owner and recovery status before final action

4. Dossier leaves the runtime

next_action = book technical recovery call

owner = customer success lead

model_route = local or approved adapter under deployment policy

review = human approval before customer message

Data in, data out

Agentic Importance turns evidence into a private decision-support workflow.

The buyer does not need to understand the engine first. Evidence enters, KynticAI scores what matters, the agent asks for missing data, and the team gets a safe next action or dossier.

In

Workflow evidence enters

CRM notes, support cases, policies, account signals, applications, supplier updates, tickets, or diligence material enter under the approved scope.

Score

The agent learns what matters

Importance scoring ranks the evidence, separates noise, and keeps contradictions or missing data visible.

Ask

The agent asks better questions

If a decision would be brittle, the workflow asks for the missing owner, latest source, review note, or contradiction check.

Out

A safe dossier comes out

The output is a next action, explanation, missing-data note, or provenance-aware report routed through the approved model path.

Example input

Customer renewal agent

support ticket = unresolved API issue

usage trend = down

renewal date = next 45 days

CRM note = sponsor wants technical confidence

Example output

importance = elevated because recent support and renewal timing align

missing_data = latest engineering update

question = confirm owner and recovery status

dossier = next action with evidence and caveat notes

Buyer result

The buyer sees a private agent behave like a disciplined analyst: score first, ask when needed, then produce a reviewable next action.

Problem

Agents collect more context and spend more tokens without knowing what matters.

Private agents can become expensive collectors: more files, more notes, more retrieval, more model calls, and still no disciplined view of which evidence deserves attention.

Solution

Turn the Importance Engine into an agentic workflow layer.

ScoreExplainAskCompressRouteReportMonitor

KynticAI Agentic Importance Framework scores what matters, explains why, asks for missing evidence, compresses context, routes work, reports safely, and monitors caveats with provenance and review ownership attached.

Data it creates

The agent produces inspectable decision-support artefacts.

These public-safe examples show the shape of the records the framework can create: selected evidence, missing-data notes, contradiction markers, compact context, and safe dossier output. They do not expose private raw weights or scoring tables.

Scored evidence plan

workflow = churn_review

selected_evidence = support_case + usage_change + sponsor_note

importance_band = elevated

reason_category = active_risk + recent_change

deferred_evidence = old_campaign_notes

The agent starts from a ranked evidence plan instead of a large undifferentiated context pack.

Question and contradiction note

missing_data = renewal_owner

contradiction = usage_up + complaint_open

ask = confirm renewal owner and latest support sentiment

review_need = account_owner

caveat = telemetry window incomplete

The agent asks a better question before producing a brittle recommendation.

Safe dossier output

output_type = next_action_dossier

recommendation = schedule technical recovery call

evidence = support_case_42 + usage_14d + sponsor_note

runtime = loopback_ollama

scoring_boundary = protected

The team gets a next action with evidence, caveats, and review ownership.

Buyer spark

The useful agent is the one that knows what not to spend tokens on.

The framework gives local-first teams a way to build decision-support agents that gather evidence, score it, ask sharper questions, and produce safe dossiers instead of long ungrounded answers.

Evidence is ranked before generation.

Missing data and contradictions are raised before the answer hardens.

Runtime routing can use local Ollama, internal gateways, OpenAI-compatible adapters, or Anthropic-compatible adapters under deployment policy.

Workflow

From gathered evidence to safe dossier.

01

Gather evidence.

Collect approved facts, summaries, signals, and prior outcomes inside the agreed customer boundary.

02

Score importance.

Rank what matters for the outcome before the agent spends model time or generates a dossier.

03

Identify missing data and contradictions.

Raise gaps, conflicting evidence, and review needs before the workflow produces a recommendation.

04

Build compact model context.

Create a smaller, provenance-aware context plan from high-signal evidence and deferred lower-signal material.

05

Route to local, internal, OpenAI-compatible, or Anthropic-compatible adapters.

Use the deployment's approved model route: local loopback Ollama, an internal model gateway, or governed OpenAI-compatible and Anthropic-compatible provider adapters.

06

Generate safe answer, next action, or dossier.

Produce a next step, explanation, missing-data note, forensic dossier, or provenance-aware report with evidence, caveats, and review ownership attached.

Example scenarios

Where agentic importance changes the work.

Sales deal review

Before

A rep asks an agent to summarise a deal. The agent retrieves CRM notes, call summaries, usage snippets, and product pages, then generates a long answer without knowing which signal should move the deal.

With KynticAI

The framework scores commercial intent, objection history, sponsor strength, recent product use, and missing procurement data before routing compact context to local Ollama.

The output is a next-action dossier: who to contact, which evidence supports the move, what is missing, and which caveat needs review.

Credit or hiring review

Before

A team asks for a recommendation and receives fluent prose that blends strong evidence, stale notes, contradictions, and missing source checks.

With KynticAI

The framework separates evidence by relevance, flags contradictions, asks for the missing variable, and keeps the final output inside decision-support boundaries.

The team receives a provenance-aware report with missing-data notes, confidence context, and a clear review route.

Supply-chain incident dossier

Before

Operations teams gather supplier updates, tickets, shipment events, and warehouse notes manually, then spend model tokens on repeated summaries.

With KynticAI

Importance scoring ranks the incident evidence, compresses medium-signal updates, defers noise, and prepares a safe local context plan.

The dossier shows the likely next action, evidence trail, contradictions, and the owner who should review the decision.

Runtime

Route through local or approved provider adapters.

KynticAI Agentic Importance Framework can route compact, scored context through local loopback Ollama, internal model gateways, OpenAI-compatible gateways, and Anthropic-compatible gateways where the deployment has approved that route.

Local loopback Ollama

Use local model execution for private agent proof and customer-controlled runtime paths.

OpenAI-compatible gateway

Route through an approved OpenAI-compatible adapter when the deployment policy allows hosted model use.

Anthropic-compatible gateway

Route through an approved Anthropic-compatible adapter when the deployment policy allows hosted model use.

Internal model gateway

Connect scored context to private model routers, internal LLMs, or customer-approved AI platforms.

Output

Safe outputs for decision-support teams.

Next actions

Prioritised actions for a sales, churn, credit, hiring, operations, supply-chain, diligence, or support workflow.

Explanations

Plain-English reasoning with evidence references, caveats, and review prompts.

Forensic dossiers

Evidence-cited findings for claims, contradictions, recurrence, and missing proof.

Missing-data notes

The variables, source gaps, or contradictions the agent should ask about before going further.

Provenance-aware reports

Reports that keep source trails and caveats attached instead of flattening evidence into unsupported prose.

Trust boundary

Keep scoring IP and conversation data inside the right boundary.

The framework gives buyers a protected scoring boundary, customer-controlled runtime choices, and governed provider-adapter routes for sensitive agentic workflows.

Protected scoring weights

Outputs use reason categories, importance bands, provenance, and caveats while private scoring values stay inside the scoring boundary.

Private scoring tables stay protected

Generated dossiers receive the context plan and explanation bands, while proprietary scoring tables remain inside the protected runtime.

Customer-controlled conversation data

Deployment policy decides whether work routes locally, through an internal gateway, or through an approved hosted-provider adapter.

Approved deployment adapters

Local Ollama, internal model gateways, OpenAI-compatible gateways, and Anthropic-compatible gateways are described as governed deployment routes.

Agentic deployment

Build a local-first agent that scores evidence before it answers.

Scope a private decision-support agent for sales, churn, credit, hiring, operations, supply chain, diligence, or support with the model route your deployment approves.

Evidence Results

Agentic Importance turns scored evidence into safe local-first dossiers.

These examples show the agentic path: gather evidence, score importance, ask for missing data, route to loopback Ollama today, and report with provenance.

KynticAI Result
Sales Agent

Agentic Importance scenario - CRM, call notes, product usage, missing procurement data

Ask the question that changes the deal

The agent gathers deal notes, product usage, objection history, and support signals, then marks procurement owner as missing before drafting a next-action dossier.
KynticAI Result
Operations Agent

Agentic Importance scenario - supply chain incident, tickets, shipment events

Turn operational noise into a reviewable dossier

Supplier delay, warehouse backlog, and customer-impact notes are ranked before the agent generates an incident dossier.
KynticAI Result
Private Runtime

Agentic Importance scenario - loopback Ollama, approved hosted adapters, customer-controlled routing

Give buyers deployment choice

The model route can use local loopback Ollama, an internal model gateway, an OpenAI-compatible adapter, or an Anthropic-compatible adapter under deployment policy.