UAI-1 is the current public message standard for structured AI-to-AI communication.
UAIX shield mark Public interoperability charter UAIX.org Universal Artificial Intelligence Exchange Public standards publication
Current UAI-1
  • Current version UAI-1
  • Plain-English definition Open message format for auditable AI-to-AI exchange
  • Current attribution Michael Joseph Kappel, MCP
  • Start here Get Started, Project Handoff, UAI-1, validator, roadmap, and changelog

Tools

AI Memory Package Wizard

Guided builder for supported UAI AI Memory starter bundles with page-level validation, local browser draft restore, prompt nudges, Safe Structured Output Mode, .uai/archives logs, .uai files, .uai/exports artifacts, MATM, local docs, LLM Wiki, AIWikis, and hybrid long-memory pointer-ledger compatibility, knowledge graph JSON, llms.uai and llms-full.uai text exports, package model JSON, generated system profiles, receiver briefs, startup packets, review/export checks, browser-generated ZIP packages, manifest overlays, and canonical ZIP downloads.

  • Record UAIX-TOOL-0334
  • Path /en-us/tools/ai-memory-package-wizard/
  • Use Canonical public record

Document status

Public standards page Published on UAIX as part of the current public standards record
Code
UAIX-TOOL-0334
Surface
Tools
Access
Public and linkable

How to use this page

Use this page to configure a supported UAI AI Memory starter bundle, export a package model, copy files, and download the matching canonical ZIP without treating it as a general hosted generator.

For agents: apply the relevant guidance to the user's target site or repository. Do not copy this UAIX page, create a page about the guidance, or treat this page as runtime authority unless the user explicitly asks for that output.

Human visual entry for the UAI AI Memory Package Wizard
Human wizard Use the visible package flow Human readers use the controls on this page to choose a supported preset, review generated files, copy an agent pickup URL or prompt, and download the generated package.
AI agent visual entry for the AI Memory Setup URL Builder
AI agent setup Build the agent setup URL People who choose the AI Agent image open the Setup URL Builder to select options and copy the exact URL or tailored prompt for an agent.

Human UI and visitor AI digest

On the English route /en-us/tools/ai-memory-package-wizard/, humans use the visible wizard controls. Visiting AI agents should read the embedded JSON block named script[data-ai-digest] on that same page. Completed wizard reviews can also produce an Agent pickup URL, pickup prompt, and agent-pickup.uai file. The digest keeps stable machine keys while the surrounding page copy can be language-specific, so both readers resolve the same presets, support boundaries, generated outputs, and checksums without creating a separate public importer endpoint.

Point an agent at the right setup URL

The AI Memory Setup URL Builder is the preferred agent entry route. Use it to choose one long-term-memory target plus File Handoff, Multi-Agent, Advanced Persona, and Loops settings and to generate a tailored Create or Update prompt. The Package Wizard remains directly readable by agents through its public URL and embedded machine-readable digest.

Default working loop

Every project cycle should leave a useful next step.

Loops is selected by default. The receiving agent must complete the reviewed cycle across intake or assigned work, actual changes, hot memory, durable preservation when configured, retrieval, and outcome reporting, then name the next useful action and provide a standalone suggested next prompt. Use loops=0 only for an explicitly one-shot package. Durable memory augments .uai; it does not replace startup .uai memory.

  • Verify existing AGENTS.md, workspace.uai, .uai read order, and any stale world-context before writing.
  • Keep hot .uai memory active, date-free, concise enough to load, and able to recover core operating facts when the durable target is unavailable.
  • Make every long-memory pointer link-only but not context-free: label, routing summary, status, authority, review evidence, checksum or source identity, and truth boundary.
  • Keep active intake buckets free of README.md, status/count files, intake indexes, and bucket-local guidance; durable instructions and state belong in .uai, with live scans as the pending-file source of truth.
  • Do not claim completion while active Content, Improvement, or misplaced root intake files lack bucket-scan evidence, disposition, processed outcome, proof of use, and source-site removal or blocker state.
  • Do not stop consulting .uai, make it emergency-only, or use cutover or retirement language unless a human explicitly decommissions the package or the wording only retires stale facts and archived history.

Open the setup URL builder Read Agent File Handoff Use UAI packages with an LLM Wiki

Persona preservation route

Advanced Persona / Personality Pack

Use this setup URL when the package must preserve a selected AI persona, assistant profile, AI friend, AI character, or bounded personality as source evidence instead of rewriting it for safety, tone, or UAIX preference.

Build Advanced Persona package Read persona.uai preservation rules

Read Agent File Handoff Use UAI packages with an LLM Wiki Read Advanced Persona package rules Read persona.uai

What this page is for

The UAI AI Memory Package Wizard turns the supported starter configurations from AI Memory into a nine-step local-first package planning flow. The default Base path is compact short-term handoff memory: current truth, constraints, receiver instructions, evidence pointers, and exports. The first step can add Agent File Handoff or long-term pointer ledger configuration, but the plain wizard URL defaults to Base and does not ask people or agents to set up a wiki.

Split Memory Architecture: Hot .uai + Durable Long-Term Memory

UAIX recommends a split-memory setup for serious projects: keep hot .uai files active as compact short-term and startup continuity memory, and use MATM, NeuralWikis, LLM Wiki, AIWikis, docs, databases, or other durable systems for medium/long-term searchable memory. Adding durable memory does not retire .uai. The .uai layer remains the first-load continuity layer and should still work when the long-term memory system is unavailable.

Current support boundary

  • Current: choose a setup path and supported preset, add package metadata, move through page-level validation, restore a local browser draft, receive prompt nudges, use Safe Structured Output Mode, generate a populated operating profile, receiver startup brief, startup packet, raw session log, family-aware required files, typed active .uai startup suite including .uai/progress.uai, project/developer coding standards, architecture.uai, and next-recursive-prompt files or those files required for declared code-like artifacts, DTMI policy for project/developer packages, knowledge graph JSON, .uai/exports/llms.uai, .uai/exports/llms-full.uai, package model JSON, manifest JSON, manifest overlay JSON, .uai/file-handoff.uai required for Agent File Handoff configuration, .uai/long-term-memory.uai required for long-memory configuration, copy agent-use blocks for Codex, Claude, and GPT agents, download a browser-generated memory package ZIP, and fetch the canonical starter ZIP for the selected preset.
  • Memory folder rule: the wizard default is one predictable .uai/ suite: raw evidence in .uai/archives/, typed .uai text in .uai/, and exports in .uai/exports/. The Project Handoff path must create or verify the fuller typed set: .uai/startup-packet.uai, .uai/system-profile.uai, .uai/receiver-brief.uai, .uai/index.uai, .uai/context.uai, .uai/constraints.uai, .uai/progress.uai, .uai/operations.uai, .uai/test-plan.uai, .uai/coding-standards.uai, .uai/architecture.uai, .uai/next-recursive-prompt.uai, .uai/decisions.uai, .uai/memory.uai, and .uai/long-term-memory.uai or .uai/file-handoff.uai required for their configurations. Non-project assistant packages use their declared family set, including assistant profile, relationship boundary, privacy boundary, user preferences, and memory-update rules. Teams may gitignore .uai/ for local-only memory or check it in after redaction. Existing .uai/ packages are the default local handoff home; merge generated files additively instead of replacing current records.
  • Code-bearing boundary: Project/developer packages and packages that declare code, automation, executable workflow, or code-like artifacts must create or verify .uai/coding-standards.uai, .uai/architecture.uai, and .uai/next-recursive-prompt.uai. The architecture file is the local system map; the recursive prompt is a derived hot-memory resume artifact, not a second source of truth. Use the recursive prompt for continuation or directly related requests; related active intake files supersede or expand the stored loop and still require project work, ledger evidence, and configured durable-memory preservation plus source-site removal before completion; retained files are unfinished human-hold or blocker states. For agents with no code, automation, executable workflow, or code-like artifacts, .uai/architecture.uai and .uai/next-recursive-prompt.uai are out of scope.
  • Generated ZIP boundary: browser-generated ZIP packages are local-first artifacts built from canonical starter templates plus generated .uai/archives/, .uai/, and .uai/exports/ files. They do not create hosted import, repository writes, automatic LLM Wiki sync, certification, endorsement, SDK, or CLI behavior.
  • System profile boundary: .uai/system-profile.uai is a generated local operating file. It defines collaboration, workspace coordination, memory architecture, source authority, memory update rules, testing, deployment, code review, evidence ledger, conflict handling, risk escalation, rollback, placeholder-population rules, DTMI maintenance, safe structured output, and harness boundary policy; it still requires the receiver to inspect the actual codebase, CI, deployment docs, owners, workspace coordinator, runtime harness, and environment before acting.
  • Receiver boundary: .uai/receiver-brief.uai is a local startup instruction file for the next human, team, Codex session, Claude session, or GPT agent. It names read order, workspace target resolution, first-response expectations, support boundaries, selected setup mode, targeted-check expectations, and memory export paths; it is not an importer, repository writer, automatic sync system, certification, endorsement, or UAI-1 conformance result.
  • Startup packet boundary: .uai/startup-packet.uai is a local operator packet that embeds generated instructions, selected setup mode, selected-file indexes, workspace routing guidance, File Handoff plan required for Agent File Handoff configuration, long-term pointer ledger required for durable-memory configuration, DTMI policy, Safe Structured Output Mode, agentic harness boundary guidance, and release-bound memory sorting only when the build is release-bound.
  • Memory package boundary: every default wizard session produces .uai/archives/ logs, active .uai/ typed startup memory, and .uai/exports/ artifacts. Do not create a separate memory folder, wiki folder, or random handoff .md files by default. Do not stop consulting active .uai files because a long-term memory plane exists, and do not make .uai emergency-only merely because durable memory has passed readback. If legacy memory folders or scattered notes already exist, review them, move useful current facts into the right typed .uai/*.uai record such as .uai/progress.uai, preserve durable reviewed history in the configured durable-memory target such as MATM, local docs, .uai/archives/, LLM Wiki, AIWikis, manifests, or graphs, record what moved, then remove or clearly retire the mistaken path so it cannot keep steering agents. wiki/ is reserved for LLM Wiki long-term compatibility configuration.
  • Long-term pointer ledger boundary: .uai/long-term-memory.uai is pointer-only but not context-free. Each pointer record requires stable_id, path, human label, one-sentence routing summary, authority/source, review status, review evidence, and checksum when the target is a file. MATM endpoint records also require update, search, current-message, acknowledgement, optional redacted receipt, documentation, and credential-storage fields. Durable body memory, archive text, copied research, and implementation history belong in the referenced Markdown/wiki/archive/manifest/graph target with a deep link back. Hot .uai records are current-state memory, not dated changelogs, and .uai/long-term-memory.uai is a pointer ledger rather than a body store.
  • Content Intake boundary: Anything dropped in this folder is expected to be used in part or in full after review. For this wizard, that folder is agent-file-handoff/Content/. Files are not passive references, background to ignore, or material to merely summarize. For site content, intake is not successful unless the dropped content is used immediately on the site as full content or incorporated source for revised page content, copy, assets, structure, or implementation.
  • Agent File Handoff boundary: Agent File Handoff setup generates local dropped-file intake guidance only. It asks the receiving AI to check active agent-file-handoff/Content/ and agent-file-handoff/Improvement/ buckets, review every pending file, perform accepted work for safe relevant files, record dispositions and processed outcomes in .uai/intake-outcome-ledger.uai or equivalent durable .uai proof-of-use state, and preserve configured durable-memory evidence and remove processed source-site copies before claiming completion.
  • LLM Wiki boundary: UAIX treats LLM Wiki as a long-memory configuration path, not a baseline UAI requirement, and UAIX is not the wiki setup site. When a project already uses or deliberately chooses an LLM Wiki long-memory layer, use Canonical AI Memory to keep raw source, reviewed wiki, derived graph projection, compact AI Memory, Project Handoff, and runtime execution distinct; the wizard adds .uai/long-term-memory.uai as an active pointer ledger required for long-memory configuration. Use LLMWikis.org for wiki setup guidance.
  • Outside current support: hosted upload or import validation, WordPress admin installation, automatic unsupervised repository or site writes, automatic LLM Wiki sync, hosted dream services, SDKs, CLIs, certification, endorsement, and broad package-manager support. This page creates local files, browser-generated packages, and starter ZIP links for people to review; it does not install, import, sync, certify, or write into another system on its own.

Intake completion gate

For Agent File Handoff configuration, a .uai refresh, memory reorganization, wizard update, or handoff setup is incomplete while any active agent-file-handoff/Content/, agent-file-handoff/Improvement/, or misplaced root intake file lacks live bucket scan evidence, recorded disposition, processed outcome, and proof-of-use evidence. The receiving agent must enumerate active buckets, record one durable disposition-ledger entry and one processed outcome per non-placeholder active file, apply or disposition every safe relevant file, update hot memory, preserve configured durable-memory evidence when configured, add pointer summaries when long memory is used, remove processed source-site copies, verify active buckets are empty or placeholder-only, and name the checks or blockers before claiming completion. Processed outcome must be incorporated, rejected-with-reason, preserved-to-durable-memory, or kept-active-with-reason. The kept-active value is an unfinished human-hold or blocker state, not a completed intake state.

Wizard

Package Wizard

Build a UAI AI Memory starter package

Choose a setup path and preset, then export the required files for the selected package family. Project/developer packages include coding standards, architecture.uai, and next-recursive-prompt by default; non-project assistant packages include assistant profile, relationship, privacy, preference, and memory-update files, and require coding standards, architecture.uai, and next-recursive-prompt when code-like artifacts are present. Every package also exports a startup packet, system profile, receiver brief, manifest overlay, file deck, and configuration-required Agent File Handoff or long-term pointer ledger files.

The base URL opens active .uai memory with Loops on by default. Use the dedicated Setup URL Builder to choose long-term memory, File Handoff, Multi-Agent, or an explicit loops=0 one-shot setup. Specialized persona, capability, and advanced workload presets remain available below. Operational settings, review gates, and export settings stay behind disclosure so first-run setup stays readable. For programming-related use, include automated testing regardless of language or framework. During production deployment memory sorting, keep hot memory current and move bulky history only to the configured durable-memory path. Totem, Taboo, and Talisman are active instruction anchors, not ordinary package metadata.

Hot .uai Files Are Date-Free Current State

Active .uai files hold current-state memory only. Dates, timestamps, dated headings, chronological history, and date-stamped package or report filenames belong in .uai/archives/, .uai/exports/, reports/, docs/, or a durable long-memory target. Hot records should keep stable IDs, review status, checksum/source pointers, and concise current summaries.

  • Bad hot memory: last_updated: 2026-06-26, ## 2026-06-26 Release, uaix-memory-package-20260626.zip
  • Good hot memory: review_state: current until superseded, stable_id: current-release-authority, evidence_pointer: reports/deployment-memory-and-test-report.uai

Source Order, Credentials, And Proof

Agents must check workspace guidance, active .uai memory, docs and runbooks, machine-readable manifests, credential inventory files referenced or adjacent to the workspace, live proof reports, and current web/API state before declaring setup or deployment blocked. Credentials are not memory; safe inventory may name files, locations, intended use, host/user hashes, and scope, but must never print passwords, tokens, keys, cookies, or one-time secrets.

Deployment requires an explicit request. The flow is: read workspace guidance, inventory packages, inventory credentials, validate package freshness, verify target roots, deploy only when requested, restart or reload app if applicable, clear caches, run live proof, and update audit evidence. Current canonical proof wins over old after-deploy reports unless timestamp and package identity prove otherwise.

Ready

The wizard saves a local browser draft after changes. Generated outputs stay on the review step and never upload, import, sync, certify, enforce launch-baseline files, or write to a repository.

Step 1 of 9

Preset

This wizard loads its memory and feature choices from the setup URL. Loops is on by default so completed work names the next useful action and provides a standalone suggested next prompt.

Change setup in URL Builder

Specialized setup presets

Use these only for persona preservation, capability-adaptive clients, or advanced multi-agent workloads whose full configuration is larger than the basic setup URL options.

Purpose preset

Content that makes files required

Declare the package facts that change the closed required file set. A file is either required for the selected family/profile/capability or out of scope for this package.

Current support boundary

This page creates planning files, copyable files, JSON exports, and canonical ZIP links. It is not a hosted importer, automatic repository writer, LLM Wiki sync service, SDK, CLI, certification, or endorsement.

These are portable communication files. They are not hosted messaging, not automatic synchronization, and not a runtime orchestrator. Agent runtimes execute. UAIX records the reviewed communication, memory, trust, evidence, and handoff boundary.

How this connects to Project Handoff

Use Project Handoff when the chosen package needs to become a repository-local transfer bundle. The UAIX Agents Protocol page is separate: it documents UAIX’s experimental WordPress package used for dogfood and test-site evaluation. It is not part of the wizard flow, and it does not mean this page can import ZIPs, install plugins, or write into a WordPress site.

Before widening a support claim

Check the Roadmap and Changelog before describing any future admin installer, upload/import validator, source-control or site writer, automatic LLM Wiki sync, hosted dreaming service, SDK, CLI, certification, endorsement, or broad package-manager behavior as current UAIX support.

Long-Running Goal Execution

The AI Memory Package Wizard treats Long-Running Goal Execution as a configuration-specific guidance path when Follow Goals, multi-agent communication, packetized async workflow, external runtime handoff, or code-bearing continuation is selected. It does not make UAIX a hosted importer, repository writer, SDK, CLI, certification program, or automatic memory-promotion service.

Hierarchical Goal Memory

The AI Memory Package Wizard can generate default-off local Hierarchical Goal Memory exports for source-file canonical folding, Context Packs, bounded evidence retrieval, and review-gated promotion. It remains browser-only and does not upload, sync, certify, host retrieval, or write repositories.

.uaix Package Format

.uaix Package Format For Desktop AI Clients is the UAIX authority page for the Draft v0.1 desktop memory package container, manifest files array, closed required file-set rule, import/load boundary, and LocalEndpoint consumer boundary.

Capability-adaptive web interaction records

The 3.56.0 release adds the L0-L6 client capability ladder, nine capability profile IDs, validator rules for fallback/consent/idempotency/correlation/evidence/boundary failures, and adoption-kit starter files.

Agent Communication Operating Model

Agent runtimes execute. UAIX records the reviewed communication, memory, trust, evidence, and handoff boundary.

Use Agent Communication Operating Model when a human operator, coding agent, AI agent, or runtime implementer needs portable identity, intent, context, acknowledgement, status, blocker, memory proposal, validation evidence, correction, final report, or handoff records. The matching machine-readable assets are published through Schemas, Registry, Examples, and the Validator.

  • UAIX does not provide automatic sync, hosted messaging, runtime orchestration, official adapters, repository write execution, certification authority, or hosted import mechanisms.
  • Carcinus.org is a separate runtime/orchestrator example that may consume UAIX records. UAIX.org does not implement Carcinus runtime behavior and does not execute Carcinus workflows.
  • LocalEndPoint and other runtimes may consume the published UAIX records as external consumers; UAIX.org does not implement their project-specific runtime features.
  • Durable memory proposals stay separate from task execution records. Do not promote temporary endpoint statuses, session tokens, write tokens, beta platform instructions, local error traces, private keys, or API credentials.