State

Current AIvan operating brief

This page is the public-safe mirror of AIvan as it exists on March 8, 2026. It keeps the current branch, proof lanes, and visible movement readable without leaking the private core.

Branch

feat/sprint12-post-v1-improvementsCurrent active workstream in the private AIvan repo.

Focus

Orchestrator v3Approved sweep covering hooks, skills, settings, MCP, memory, and multi-AI behavior.

Surface

aivan.ivanm.devDedicated public lane under ivanm.dev, deployed through Vercel with Cloudflare DNS.

Current signals

The public site tracks real operating movement.

These are the visible numbers that keep the brand tied to shipped behavior instead of generic AI copy.

24plugins enabled

Configured in AIvan settings as part of the working orchestrator surface.

40+subagent types

Routing already spans explore, plan, bash, review, feature, and execution roles.

10automation hooks

Hooks keep learning, status, and cleanup moving without manual babysitting.

4provider lanes

Gemini, Claude, OpenAI, and local OpenAI-compatible endpoints are all supported.

4security modes

Confirm, trust-confirm, sandbox, and unrestricted execution are part of the product posture.

1public site lane

The website ships from its own repo and domain instead of hiding inside the portfolio stack.

Working now

Proof lanes already visible in the repo

Natural-language transformsnode ./bin/dev.js transform ./examples/orders.json "Sort by total descending and keep top 10"

Intent becomes executable local data work instead of a one-off answer.

Structured-data transforms already run today for JSON, JSONL, and NDJSON inputs.

Provider routingnode ./bin/dev.js transform ./examples/events.jsonl "Group by region" --provider claude

Provider choice is a runtime lane, not a redesign project.

The same task can move across Gemini, Claude, OpenAI, and local OpenAI-compatible endpoints.

Workflow lanenode ./bin/dev.js workflow --help

Workflow automation is already a first-class command surface.

The direction stays DAG-based task execution rather than a pile of prompt snippets.

Boundary

Public-safe by design

Public: product shape, capabilities, command lanes, architecture posture, release log, and deployment status.
Private: research artifacts, patent work, internal prompts, sensitive memory, and detailed orchestration internals.
Rule: show confidence and specificity without leaking the materials that create AIvan's edge.

Operating lanes

The current shape is broad, but still legible.

Transform engine

Structured inputs, generated execution plans, local pipeline execution, and practical reshaping already exist in the CLI.

Natural-language input mapped to executable transforms
Core operators include filter, sort, limit, group_by, and pivot
Output paths and in-place writes are supported by the command flow

Provider routing

AIvan is model-agnostic by design so quality, latency, or quota shifts do not collapse the whole system.

Gemini, Claude, OpenAI, and local endpoints
Provider manager and routing layer in the core runtime
Quota and performance tracking already exist in the orchestrator

Memory system

Hot memory, cold store topics, learning journals, patterns, and error entries keep the next session from starting with amnesia.

MEMORY.md is the hot cache
.orchestration stores durable topics, patterns, and decisions
Hooks promote stable lessons after meaningful sessions

Execution surfaces

The CLI, workflow system, scaffold lane, visualizer, and voice experiments all sit under one product identity.

Workflow and scaffold commands are already shipped
Visualizer code and assets are part of the active repo
Voice remains prototype-oriented but operationally real

Selective release ledger

March 8, 2026

AIvan website moved into its own release lane

The public surface now ships from a dedicated Next.js repo at aivan.ivanm.dev instead of staying coupled to the main portfolio codebase.

March 8, 2026

Orchestrator v3 full-sweep design approved

Six parallel audit lanes identified infrastructure, content, and feature upgrades across hooks, skills, settings, MCP, cold store, and multi-AI behavior.

March 7, 2026

Auto-push and session-end learning became operational

AIvan now persists orchestration and research updates with less manual cleanup, strengthening the compounding knowledge loop.

March 6, 2026

Seed specification locked the product shape

The current product direction was codified through an Ouroboros interview-to-seed flow covering transforms, scaffolding, workflows, voice, visualization, and security.

March 1, 2026

Orchestrator v2 established the current operating model

Hot and cold memory, hook orchestration, plugin skills, MCP usage, and multi-AI routing were formalized into the core system architecture.