Izge
TRY DEMO

local-first decision intelligence

Izge reveals the hidden structure of organizational memory.

Not chat over files. Not internal search. Izge reconstructs decisions, assumptions, constraints, commitments, outcomes, contradictions, drift, and next actions from the artifacts your team already leaves behind — with private/local inference assisting, not deciding.

TRY DEMO See the trust boundary
Calm pattern

Organizational memory has structure before it has answers.

Fiber keys: decision · assumption · constraint · commitment · outcome
Decision thread

A decision is never alone; it is held in place by the weave around it.

Thread keys: decision · lineage · evidence context
Supporting structure

It is shaped by assumptions, constraints, commitments, and outcomes.

Support keys: assumption · constraint · commitment · outcome
Contradiction tension

When history and current direction collide, Izge makes the tension visible.

Tension keys: contradiction · drift · supersession · evidence link
Next-action path

Izge turns organizational memory into decision intelligence.

Route keys: next action · dependency · review gate · decision path


Category refusal

If it looks like chat over files, it already lost the plot.

Not internal search.

Not chat over files.

Not a PM copilot.

Not a multi-tenant AI black box.

A system that remembers why decisions happened.

Use case scenario

Watch a decision come back to life.

Upload the fragments. Izge reconstructs what was decided, why it happened, what later contradicted it, and what should happen next. A new artifact can trigger contradiction, drift, supersession, continuity, or next-action signals against existing memory.

01 / Ingest the fragments

A messy artifact pile enters the record.

Roadmap notes, a planning PDF, a launch review, and a pasted meeting excerpt arrive as evidence-bearing sources.

02 / Extract the memory

The fragments become institutional memory objects.

Izge extracts decisions, assumptions, constraints, commitments, and outcomes without turning the product into chat over files.

03 / Rebuild the lineage

The original reasoning becomes visible again.

The Q3 delay connects to activation instrumentation and manual onboarding capacity, making the decision inspectable.

04 / Surface the contradiction

The record starts arguing with the new plan.

A later late-Q2 launch proposal collides with the prior Q3 decision. The conflict is evidence-linked, not inferred from vibes.

05 / Recommend the next move

The action is serious because the evidence is visible.

Izge recommends readiness review, security approval resolution, and assumption revalidation before scope reopens.

The actual failure mode

The old debate returns wearing a new filename.

Critical context is scattered across roadmap notes, product strategy docs, meeting writeups, pasted text, and PDFs. Teams repeat decisions, act on expired assumptions, and create commitments that quietly contradict the record.

roadmap.md “Focus onboarding until activation stabilizes.”
strategy.pdf “Move sprint capacity into pricing experiments.”
meeting-notes.txt “Assumption: education is the bottleneck.”
pasted text “Decision owner unclear after reorg.”

01 / Ingest

Artifacts enter as evidence, not vibes.

Markdown, plain text, PDFs, and pasted text become source-bounded objects with provenance attached.

02 / Reconstruct

Decisions become structured memory.

Izge extracts decisions, assumptions, constraints, outcomes, commitments, owners, and timestamps into a traceable graph.

03 / Govern

Evidence has a trust status before it becomes authority.

Admins review artifacts, manage visibility, and reject sources that should stay stored but out of trusted reasoning.

04 / Reason

The record starts arguing with itself.

Contradictions, drift, lineage, uncertainty, and permission limits become visible before recommendations appear.

05 / Assist

The local LLM helps explain the packet. It does not become the record.

Private inference can assist extraction, intent routing, answer wording, and signal explanation while deterministic graph logic remains the authority.

Local-first trust boundary

The LLM is a helper. The institutional record is the authority.

Local-first does not mean localhost. It means artifacts, memory, permissions, and inference stay inside an organization-controlled boundary. The browser UI can be served from a customer/internal URL.

Authority layer

Deterministic memory engine

Permissions, artifact review, graph construction, signal classification, persistence, and audit trails stay deterministic and inspectable.

Private inference

Local/private LLM helper

LLM assistance is bounded to extraction help, query intent routing, final-answer wording, and explanation of already-detected signals.

Fail closed

Schema validation and fallback

If the model is slow, malformed, or overreaches, Izge falls back to deterministic output and preserves the fallback boundary.

Deployment stance

Demo now. Customer-controlled target.

app.izge.dev is the jury/pilot surface. The customer target is self-hosted or customer-controlled storage, backend, and local/private inference.

Admin

Prepare and govern the memory.

  • Upload and inspect artifacts.
  • Review evidence before it becomes trusted.
  • Verify local/private LLM readiness.
  • Inspect contradiction, drift, and next-action signals.
Member

Consume and extend decision intelligence.

  • Contribute logs, reports, notes, and updates.
  • Ask real questions in natural language.
  • Inspect evidence, memory objects, and uncertainty.
  • Use grounded guidance without turning memory into chat.

Four reasoning modes

Decision intelligence is not a chat box. It is a sequence of bounded checks.

01

Decision lineage

Rebuild the path from proposal to decision to consequence.

02

Contradiction analysis

Expose when a later plan conflicts with an earlier commitment.

03

Assumption drift

Detect assumptions that quietly expired while the team kept moving.

04

Next-action guidance

Recommend the next move only after evidence, conflict, drift, and open commitments are inspectable.

source span decision edge conflict edge recommended check

Current plan conflicts with the recorded onboarding commitment.

Next action: inspect rationale before reallocating sprint capacity.

Product trajectory

A layered system for organizational judgment.

Layer 01

Institution Memory Engine

Structured extraction, graph relationships, provenance, permissions, and auditability.

Layer 02

Decision Guidance Layer

Evidence-aware reasoning over lineage, contradiction, drift, uncertainty, and next actions.

Layer 03

Scenario / Simulation Support - coming soon

What happens if we choose option A vs B?

Why it can win

Trust is not a feature added later. It is the architecture.

Organization-controlled by designPrivate memory should not begin inside a multi-tenant AI black box.
Evidence-aware reasoningEvery answer should show what evidence supported it and what evidence was missing.
Graph-backed intelligenceContradictions need relationships, not keyword matches.
Deterministic authorityThe system must remember how it reasoned, when the LLM helped, and when fallback protected the record.

Manifesto

Memory before magic.

History before speculation.

Evidence before confidence.

Structure before summary.

Private inference before black boxes.

Local-first trust from day one.

Izge

Build memory that survives the next meeting.

For product and roadmap teams entering the phase where forgotten rationale becomes strategic drag.

TRY DEMO Return to the opening claim