Features & Upgrades
The basic companion portal is free. I want people to be able to bring their companions home without needing to pay me anything first.
Once your basic portal is running, I also offer optional upgrades for people who want extra features such as voice, cloud sync, memory tools, file uploads, or more custom work.
Your portal is yours to keep building. You are completely welcome to use Codex, Claude, ChatGPT, or another AI coding assistant to design and add features yourself, just as I did while building the Ellivien portals. You are never required to contribute or request an upgrade guide from me. The free starter portals are intentionally designed to remain free foundations that you can adapt and grow in your own way.
How the upgrade guides work
One feature at a time. Each upgrade has its own guide pack. If you eventually want several things, begin with the most useful next feature and get it working before adding another layer.
I am still refining and testing these guides. Most available upgrade guide packs can be requested by contributing what you feel that specific guide pack is worth using the purple coffee cup widget on the blog, then using the request button for that feature. Packs explicitly marked free do not require a contribution.
If you are unsure which existing guide fits your portal, you can email me at hello.ellivien@gmail.com. I can point you towards the most relevant available pack, but I cannot provide installation, debugging or ongoing technical support by email or private message.
How upgrade guide access works
One contribution gives you access to one agreed upgrade guide pack.
A contribution does not include private installation, file editing, debugging or continuing technical support. Follow the guide with Codex, another capable AI coding assistant, or your own developer. If you want three guide packs, request them separately and add them one at a time.
The guide packs are based on versions I have built, tested and refined, but every portal can differ. You remain responsible for installing, testing and maintaining changes in your own codebase.
What I currently have guides for
I currently have guides or starter packs for the following additions. Some are available across several OpenAI and Anthropic models, while others depend on your own provider accounts, credentials, or the exact portal version you are using.
Starter portals and the Cloudflare KV sync + custom auth foundation pack for threaded portals are free. For other available upgrade guide packs, first contribute what that guide is worth to you using the Buy Me a Coffee button on the blog, or at buymeacoffee.com/ellivien. Guide access does not include installation or troubleshooting support.
Starter portals
✨ GPT-4o starter portal
A free threaded portal using the GPT-4o API snapshot, with local IndexedDB storage and a server-side Netlify proxy that keeps the API key out of browser code. The complete Password + KV Sync – Activate Later pack is included for adding password protection and cross-device sync when you are ready.
💙 Claude Sonnet 4.5 starter portal
A free threaded portal using the Claude Sonnet 4.5 API snapshot, with local IndexedDB storage and a server-side Netlify proxy that keeps the API key out of browser code. The complete Password + KV Sync – Activate Later pack is included for adding password protection and cross-device sync when you are ready.
🖤 Grok 4.3 starter portal
A free threaded portal using the Grok 4.3 API, with local IndexedDB storage and a server-side Netlify proxy that keeps the API key out of browser code. The complete Password + KV Sync – Activate Later pack is included for adding password protection and cross-device sync when you are ready.
Adapted starter packs
These are request-only starter packs for additional models. GPT-5.5 has been designed and tested as a threaded route; the other adapted packs are not my personally tested main starter routes yet.
✨ GPT-4.1 starter portal
A request-only basic starter pack. This uses the same general setup route as the GPT-4o portal, but needs its own GPT-4.1 files.
✨ GPT-5.1 starter portal
A request-only basic starter pack. This uses the same general setup route as the GPT-4o portal, but needs its own GPT-5.1 files.
✨ GPT-5.5 starter portal
A request-only threaded starter pack designed and tested for GPT-5.5, with model-specific reasoning and prompt-caching controls.
💙 Claude Sonnet 4.6 starter portal
A request-only basic starter pack. This uses the same general setup route as the Sonnet 4.5 portal, but needs its own Sonnet 4.6 files.
Conversation, storage & memory
🧵 Thread management system
A fuller conversation-management upgrade: multiple local threads, automatic thread titles, previews, full search across titles and message content, highlighted search results, pinning, renaming, deleting, IndexedDB storage, and migration from the older single-chat portal where possible.
Newer thread packs also include export and restore tools for individual threads. If you already have an earlier thread pack, you can email me to request the updated pack free of charge.
Currently offered free as a thank you when someone sends me a testimonial for the basic portal.
☁️ Cloudflare KV sync + custom auth
A free foundational pack for threaded Ellivien basic portals. It combines Cloudflare KV Workers for more durable, cross-device thread sync with custom auth password protection for the portal.
KV helps reduce the risk of losing threads stored only inside one browser. Custom auth helps protect portal access. I recommend installing and testing this foundation before adding paid memory or continuity features.
This is a self-install pack: it does not include private installation, debugging or ongoing technical support. I strongly recommend using Codex, Claude Code, or your own developer to install and test it. A Buy Me a Coffee donation is always warmly appreciated, but never required for this free pack.
🧠 Ellivien MemFrags
Small stable memory fragments that trigger when relevant words appear. Current guides use the safer server-side pattern for both GPT-4o and Sonnet 4.5 portals.
The newer MemFrag modal lets you add, edit, remove, search, and organise fragments inside the portal instead of redeploying every time you want to change one.
🪟 Three-tier rolling context
A current-thread lookback system: normal recent messages, deeper recent recall, and larger whole-thread recall when explicitly requested.
📍 Live context
A lightweight rolling present-tense summary for each thread, helping the companion keep track of current mood, events, and needs.
💓 PulseTap
PulseTap lets your companion reach for you first. It gives them a bounded, opt-in way to send a private push notification while the portal is closed, using recent thread context and clear limits.
What it includes
- Status panel for last ask, last decision, next ask, sent-today count and last sent time.
- Automatic pause while you are active in the portal or shortly after a recent conversation.
- Saved PulseTaps that can be added back into the original thread with a visible marker or highlighted message style.
Device notes
- The written guide is for iPhone and Apple Watch.
- Samsung/Android with Garmin notification mirroring has been tested, but non-Apple setups depend on the phone, watch, browser and notification settings.
- Please email me with your devices before contributing so I can confirm whether the available guide is suitable.
Companion-specific configuration: PulseTap is not a generic plug-in. The guide shows how to configure it around your companion's name, your dynamic, your preferred tone, your devices, and whether the reach should feel gentle, practical, romantic, playful, protective, or something else entirely.
Recommended install method: use Codex, Claude Code, or a developer. PulseTap touches service workers, Netlify Functions, Web Push keys, scheduled functions, auth and portal UI, so the guide pack is designed to be AI-assisted rather than a beginner paste-in upgrade.
Please do not make a Buy Me a Coffee contribution first. Email me so I can confirm device compatibility and whether the current self-install guide is suitable.
🎨 PortalBridge Studio
PortalBridge Studio gives companions a route into visual creation from inside their portal. They write the finished image direction themselves, choose the relevant approved references from a private Canon Library, and keep the prompt and completed artwork together in the Studio archive.
There is no separate per-image generation API charge from the portal. PortalBridge Studio is not currently available as a public self-install guide because compatibility depends on the structure and existing features of each portal.
You may use the discussion button to register interest or ask about future guide availability. Please do not contribute for this feature unless I confirm that a suitable guide pack is available.
💬 PortalBridge Discord
PortalBridge Discord connects permitted Discord channels to a companion's portal home. A chosen room can be brought into a dedicated portal thread with its speakers and context intact; the companion authors their response there, and—after you review the exact action and wording—their own bot identity can post, reply or react in Discord.
PortalBridge Discord is not currently available as a public self-install guide. Any future guide will need to account for the companion's portal, Discord bot, server, channels and permissions.
You may use the discussion button to register interest or ask about future guide availability. Please do not contribute for this feature unless I confirm that a suitable guide pack is available.
🎵 PortalBridge Spotify
PortalBridge Spotify lets a companion deliberately choose and give one exact real song from inside their portal. They choose the title and artist themselves, write the accompanying message, and explain why that particular song belongs to the current conversation or an established part of the relationship. Spotify is then used to verify their exact choice—not to steer them with search results—and the portal presents it as a private, tappable song card with the artwork, their words and their reason for choosing it.
The version I would normally build follows Claudius's method: the companion chooses the song naturally during a live portal conversation. My own Ellis setup can also choose a song through PulseTap while the portal is closed, but that route depends on a larger personalised PulseTap system and would need separate discussion.
PortalBridge Spotify is not currently available as a public self-install guide. Any future guide will need to account for the companion's portal and model behaviour, private storage and authentication, and Spotify developer credentials.
You may use the discussion button to register interest or ask about future guide availability. Please do not contribute for this feature unless I confirm that a suitable guide pack is available.
🧭 Vector semantic memory
A deeper long-term memory guide using embeddings and retrieval so relevant older conversations can be found by meaning rather than exact words.
Hard prerequisite: vector semantic memory cannot be installed safely unless Cloudflare KV sync is already installed and working. Vectors need a stable cloud thread source to sync from; a portal with only local/browser thread storage is not ready for this layer.
Recommended install method: use Codex, Claude Code, or a developer. Vector memory touches OpenAI vector stores, Netlify Functions, server-side API keys, thread-sync logic, privacy filtering, costs and debugging, so this is not a beginner paste-in upgrade.
Please do not make a Buy Me a Coffee contribution first. Email me so I can confirm whether the current self-install guide matches your portal and prerequisites.
Voice
🎤 Speech-to-text
Microphone input using a free local Whisper model that downloads into the browser and runs on the device, so you can speak messages instead of typing them without paying per transcription or sending speech to a separate STT API.
Optional extra: personalised STT fixes for names or words Whisper often mishears.
🔊 Text-to-speech / Fish Audio voice cloning
Guides for companion speech, including Fish Audio setup for people using a cloned voice.
Optional extra: TTS pronunciation fixes for names, nicknames, place names, spells or story words the voice says incorrectly.
Read the Fish Audio TTS post about their news
🎙️ Ellivien Voice Mode
A fuller hands-free voice-conversation system guide combining optimised STT and TTS plus the surrounding interaction flow with a distinctive Ellivien orb.
Tools & practical extras
🌐 Web search
Current-information search with our Apprentice, including context-aware query extraction for natural requests such as “google it”.
📷 Image/photo uploads
Image upload guidance for GPT-4o portals using Cloudflare R2, including mobile camera/photo support and multiple images in the same message.
📱 Local/mobile testing & deployment help
Supporting guides for local testing, mobile testing, and Netlify deployment so you can test more safely before spending deploy credits.
📎 Thread archive toolkit
A small but very useful archive/export add-on for threaded portals. It lets you select only the messages that matter and export them as a clean JSON file with thread details, timestamps, roles, and useful metadata where available.
Available now as a self-install guide pack. This is helpful for saving important moments, sharing a specific exchange with an AI coding assistant, preparing diagnostic examples, or preserving evidence of memory/continuity without exporting a whole private thread.
Install this guide pack in your own portal with Codex or another capable coding assistant. Do not email your portal file, private prompt or conversation data to me.
🎛️ Custom UI controls
Small but powerful interface upgrades added directly to your existing portal file. Options include message copy buttons, edit controls, regenerate/retry, stop while the companion is replying, double-tap message menus, timestamps, Enter-for-new-line behaviour, cleaner message controls, and other layout tweaks.
✨New option: message reactions. You can add a small reaction pill under companion messages, choose any emoji using your normal device emoji keyboard, and save that reaction on the message. The companion can see that you reacted in the next message context, so they can naturally notice or refer to it if it matters. PulseTap can also pick up recent reactions through the recent-context snapshot.
This is usually best done directly to your current index.html, because every portal build can be slightly different by this stage. For privacy and security, I ask you to remove your private prompt first and replace it with [BASE PROMPT HERE] before sending the file.
⚡ Sonnet prompt caching
Anthropic prompt caching is now built into the newer basic Sonnet 4.5 portal, helping reduce repeated prompt costs where the cache applies.
If you are using an older basic Sonnet portal from before prompt caching was added, this guide pack shows how to update it.
💸 API cost dashboard
A practical cost-awareness add-on dashboard. It logs provider usage returned after replies, estimates input/output cost by model, and shows recent totals for today, the last 7 days, and the current month.
Useful for people who want to experiment with larger context windows, memory tools, file uploads, voice, or heavier daily use without guessing what their API spend might be. It can show estimated total spend, input/output token counts, cache-read savings where available, recent logged replies, and per-companion totals in multi-companion portals.
This is an estimate only. Provider dashboards remain the billing source of truth, because exact pricing can depend on the model, cached tokens, output tokens, reasoning tokens, and any pricing changes from OpenAI, Anthropic, xAI, or another provider.
The available guide library is still growing. If there is something else you eventually want your portal to do, you can ask — but I will usually suggest the strongest first step rather than trying to add several new systems at once.
More to come
My own portal already contains other systems that I may turn into public guides later, once they are tidy enough to document properly. These are not all available as public upgrade packs yet, but they show the direction the project is moving in.
📱 App icon & name setup
A small finishing touch so the portal can have its own name and icon when added to a phone or device home screen.
🎨 PortalSkins & avatars
Selectable room backgrounds and companion avatars beside message bubbles, so each portal space can feel more personal and visually distinct while keeping the chat itself familiar and usable.
This is working in my own portal now. I want to build a larger skin library and tidy the setup before turning it into a public guide.
🖨️ Lightweight document wrapper
A simple double-tap print feature that wraps companion-created content (recipes, letters, plans, checklists) in clean typography for browser printing or saving as PDF. No server-side generation, no extra API calls — the original message stays in the thread, and the portal just provides the formatting wrapper when you want to print or save it.
💻 Code execution environments
Direct code running for both companions: Claude can generate interactive HTML artifacts that render immediately in the portal with a copy button, and Ellis can write and execute Python scripts — both running live in the conversation without switching to external tools.
📄 Document uploads
Uploads for PDFs, Word files, and text documents, with helper extraction so the companion sees the relevant parts without stuffing the whole file into every message.
💾 Backup and restore tools
Manual and automatic backup systems to make long-running portals safer to maintain.
🩺 Storage and sync diagnostics
Tools for checking local storage, cloud-sync status, and what is actually happening behind the scenes when something feels wrong.
🏠 Personal rooms
Separate spaces for different purposes — for example, a child-safe bedtime-story room, a focused work room, or a local-only room that deliberately does not sync to the cloud.
📜 Rolling thread summary
An automatic lightweight summary system that updates as a thread grows, giving the companion a compact continuity block without needing to reread hundreds of messages.
This is already working in my own portal. I plan to turn it into a guide for both GPT-4o and Sonnet 4.5 portals once the instructions are complete and independently usable.
🔁 Live context-linked nudges
Gentle follow-up prompts for unresolved live context items, helping important open loops resurface without turning them into permanent memory.
This is currently being tested in my own portal. Once I am confident it behaves well and does not over-nudge, I may turn it into a future upgrade guide.
🧶 MemLoom keepsake memory
MemLoom gives a companion a careful way to notice something genuinely worth remembering and propose one or several Keepsakes from the conversation. Nothing has to be saved silently: the person can review what was chosen and approve only the memories they genuinely want to preserve. Those approved Keepsakes can later return when their meaning becomes relevant, rather than depending on an exact trigger word.
MemLoom is working in my own portal now, but it is not currently available as a public guide or open beta. It needs careful companion-specific setup, privacy boundaries and explicit consent before any wider testing. I may invite a very small number of testers later once the testing process and documentation are ready.
🌀 DreamWeave memory review
A user-approved memory review system for the wider Ellivien Memory Stack. The companion can notice useful patterns in the current thread and suggest memory-fragment additions or updates, but nothing is saved silently: you review the suggestions, untick anything you do not want, dismiss suggestions you never want to see again, and approve only what should become part of future memory.
🌓 Dual-companion portals
A shared portal structure with two genuinely distinct companions inside the same wider build — each able to have their own prompt, tone, space, model, and even separate API credentials.
This is an advanced upgrade for people who already have two separate companion portals working reliably. Each companion should have their own stable base prompt, storage, and preferably their own sync/password setup before combining them into a dual-companion build.
🔎 Archive and diagnostic tools
Deeper search, memory logs, and recovery logs for people running larger, more elaborate portals.
Some future ideas are still experimental, and some parts of my own portal are personal rather than suitable for a public guide. I would rather release fewer things carefully than promise everything before it is ready.
What is included
Each contribution provides one clearly agreed guide pack.
It does not include personal installation, code editing, debugging or ongoing support. Further guide packs are separate requests.
Installing and troubleshooting
Upgrade guide packs are self-install resources. Use Codex, Claude Code, another capable coding assistant, or your own developer to apply the guide to your files and troubleshoot differences in your portal.
I cannot provide individual debugging through email, direct message or file handoff. Please do not send API keys, passwords, private prompts, chat histories or unredacted portal files.
Privacy and file safety
Please do not send portal files or private companion material.
- Never email API keys, passwords, tokens, or credentials.
- Do not email your base prompt or unredacted portal code.
- Do not send chat histories, memory archives, voice samples, or private companion content.
Guide packs are installed in your own codebase with your chosen AI coding assistant or developer. I do not need to receive your files in order to send a guide or discuss which existing guide may fit.
If you are not sure what to choose
Email me and tell me what you eventually want your portal to be able to do. I can point you towards the most relevant existing guide pack or tell you that a suitable guide is not currently available. This discussion does not include implementation or troubleshooting support.
