Complete Guide to Perfect Chat Memory · Zero Context Loss · Infinite RP Length

Two complete export workflows plus one Legacy fallback preserve your story without corrupting the AI’s perception.
The core principle: Separate History from Storytelling.

Once a long chat fills its context window, the model loses access to the beginning, starts filling gaps, and drifts from the character’s voice. Think 65,536 tokens (≈65k) for a 64k window, or 128k for a larger setup. Standard auto-summarization usually flattens emotional depth and physical micro-details.

The fix is to treat your story like a game with a save file. Below is how — in plain language anyone can follow.

ACTIVATION
═════ △ ═════
HOW TO START

📋 Where do I get the prompts?
Right here in this GUIDE’s chat. Choose the matching message below and copy its full prompt.

🎮 ANY BOT
Messages 1 and 2 — Tool 1 and Tool 2.
For any RP bot, including characters with complex psychology, custom mechanics, or their own HUD. The prompts follow what your chat actually shows.

🔬 MY BOT
Messages 3 and 4 — Tool 1 and Tool 2.
For bots created by me. Use this lane when the bot is one of my releases; it preserves the visible mechanics built into those bots alongside the played story.

Tool 1 — Save State: records the established story and exact present state for Chat Memory.
Tool 2 — Bootstrap Generator: moves the story into a new chat to free context, carrying a State Snapshot and a literary Scene Start.

⚠️ Golden Rule — read this once

Export the played log and run the matching tool in an external long-context model.
This is now the standard workflow for both Tool 1 and Tool 2 on PC and Android. Keep extraction outside the played story; direct in-chat execution remains available only as the Legacy fallback below.

METHOD 1 · BEST QUALITY · PC + ANDROID
═════ △ ═════
The External Archiver

Let a "smart" AI read and summarize the whole chat for you.

How it works: Export the chat → send it to an external AI with the Save State (Tool 1) prompt → paste the result into Chat Memory.

⚡ Step 1: Get the chat out of the site
🖥️ DESKTOP / PC · CHROME / CHROMIUM
Open the desktop exporter, install it in your compatible browser, then export the JanitorAI chat.

📥 DESKTOP EXPORTER · OPEN

📱 ANDROID / MOBILE · FIREFOX
1. Install Firefox for Android.
2. Open the JanitorAI Chat Downloader, Reader & Exporter page inside Firefox.
3. Tap Add to Firefox, then confirm with Add.
4. Open Firefox’s Extensions list and confirm that the exporter is installed.

Used by Method 1 and Method 3. Method 1 prepares the log for Tool 1 / Chat Memory. Method 3 prepares it for Tool 2 / Bootstrap. Method 2 is the Legacy in-chat fallback and does not use the exporter.

💾 Step 2: Save it as plain text
Export as .txt or .json — no formatting garbage.

🧠 Step 3: Send it to a model with a long context window
Paste the Save State (Tool 1) prompt, then attach or paste your full chat log in the same request. Include your previous save when you have one.

Services for Summarization:
• Grok
• Gemini · Google AI Studio
• Qwen
• DeepSeek

Choose enough context for the whole job: the log, the prompt, and the completed report. Paste Tool 1 first, then your log; the AI returns a ready memory save.

💾 Step 4: Put it back into memory
Copy the generated report → in Janitor go to chat settings → Chat Memory → paste it in.

Replace the older Chat Memory with the new consolidated report once it carries the still-active facts from your previous save. Keep the exported log as your backup.

Pros and Cons

➕ Maximum precision • Clean perception • Backup included • Works on PC and Android
➖ Requires installing a compatible browser extension once.

METHOD 2 · LEGACY FALLBACK
═════ △ ═════
The Quick Compressor · Inside the RP Chat

How it works: Paste Save State (Tool 1) as your next message → bot replies with the summary → copy it to Chat Memory → delete both messages.

Run it while the chat still follows the story cleanly.

Updating an existing memory? The small, useful merge tip from @daepiksigma is built into the updated prompts: combine only new information and supported changes with the previous save, keep its still-active history, and return one complete updated memory. Make the previous save available to the model; there is no extra line to append each time.

Pros and Cons

➕ No exporter required • Fast emergency fallback
➖ It can read only the context still available to the RP model, and the extraction request temporarily enters the played chat. Use Method 1 for the complete exported history.

METHOD 3 · A NEW CHAT TO FREE CONTEXT
═════ △ ═════
The Hard Reset / Bootstrap

How it works: give the matching Bootstrap Generator (Tool 2) the played log and any previous save → get one completed bootstrap → paste it into a new chat with the same bot.

The new chat carries your preserved story state instead of the entire old message history, freeing context for the next stretch of RP.

1. Generate the bootstrap
Use the desktop or Android exporter from Method 1. Give an external long-context model the matching Tool 2 prompt, the exported log, and your latest Tool 1 save when available.

2. Move the complete result
Copy the whole output, including SESSION BOOTSTRAP → STATE SNAPSHOT → SCENE START. Open a new chat with the same bot and paste it there. Keep the bot’s required runtime and Lorebook attached.

3. Continue from the Scene Start
Reply to its new scene beat. If the bot is asked to render that opening, play it once; continue after anything already accepted or played. Your choices and response remain yours.

Pros and Cons

➕ A fresh context window for the next stretch of RP
➕ Preserved history, exact scene physics, and a natural literary opening
➖ Requires moving the finished packet into a new chat.

Optional: request an in-place re-anchor to restate the saved state inside the existing chat. To free context, use the new-chat workflow above.

💡 Pro Tip — The Hybrid Approach

Use a fresh export with Tool 1 for a memory update roughly every 50–80 messages.
Use the export plus the latest Tool 1 save with Tool 2 roughly every 200–400 messages, or at a major arc break.
Keep Method 2 only as a Legacy fallback when export is unavailable.

Watch context usage first; message counts are a starting rhythm. On a 128k setup, if you are below ≈65k tokens and the story is coherent, you can simply keep playing. As the window fills, save and prepare the next bootstrap. On a smaller window, move that checkpoint earlier.

🤔 What about slow-burn and minMessages?

“Will a fresh chat erase the progress we earned?”
The save carries the relationship’s path, revealed information, current boundaries, promises, and consequences. You resume at the emotional point you actually reached.

Story progress and message gates have separate jobs. The snapshot preserves what has already happened; the original Lorebook or script still controls when it delivers an entry. Keep those systems attached when restarting.

UPDATED MEMORY PROMPTS
═════ △ ═════
What the Save Protects

The tools preserve your chat as a playable state: what happened, what changed, what was revealed, what remains active, and the exact point where play stopped. They carry visible mechanics when present and compress the wording without discarding continuity.

🌍 Chat-established world continuity — New or activated facts, demonstrated rules, clarified history, and lasting changes created by this chat.

🧬 Current physical continuity — Position, distance, contact, posture, gaze, clothing, injuries, active effects, and unfinished actions. Lasting body changes remain distinct from the current pose.

🏠 Locations, environment & territory — Familiar homes and workplaces keep their distinctive layout and identity. Changed rooms, damaged or repaired structures, passages, access, and control keep their state when you return.

🗣️ Voice & emotional state — Current tone, speech changes, emotional residue, and character-specific language established in play.

🔗 Objects, promises & consequences — Important objects and their state or location; commitments, debts, costs, limits, and unresolved outcomes.

🌀 Symbols, motifs & relationship trajectory — Recurring images and gestures keep their earned meaning. Decisive shifts remain connected to the current dynamic.

↩ Callbacks, knowledge & visible mechanics — Exact callback lines, who knows or believes what, and supported status values from the chat’s own mechanics.

🏠 Return to the same house

Easier location recall is built into all four updated prompts. Your home, the character’s home, and the workplace each keep their own established identity and the distinctive layout or features that matter on return — even when nothing has been rebuilt.

Location registry idea suggested by @Naturally_Blonde.

“A stands by the door; B sits on the sofa” belongs to the current scene.
“The spare room is now a clinic; the west stair is sealed; Mara controls access to the cellar” belongs to persistent space.

All four prompts carry location anchors and lasting changes together in 【LOCATION REGISTRY — PERSISTENT ENVIRONMENT】. A scene change keeps them in place. A repair updates the repaired part. If time passes without another observation, the save keeps the last-known condition.

Movable objects keep their condition and location in 【OBJECTS / PROPS LEDGER】: the emergency key now lives in the kitchen drawer. A separate consequence for the place — who can open the cellar — belongs with that place.

Save the details that can change a later scene. Ordinary furniture does not need a room-by-room inventory.

COMPRESSION & EVIDENCE
═════ △ ═════
The Core Rules

1. CONTINUITY > BREVITY. Keep unique details with future causal, relational, physical, voice, or callback value. Active consequences retain their meaning even when the wording becomes shorter.

2. SAVE WHAT THE CHAT ESTABLISHED. Record enacted events, attributed claims, plans, and interpretations as their own kinds of evidence. Preserve who knows what. Unavailable card or system content stays outside the extraction.

3. ENGLISH OUTPUT; EXACT QUOTES STAY EXACT. All four prompts produce English memory, bootstrap content, Scene Starts, and headings, regardless of the RP’s language. Necessary exact source wording is marked VERBATIM and preserved unchanged.

4. PRESERVE HOW THE STATE WAS EARNED. Keep the decisive relationship changes and their causes. Copy visible mechanics with their actual owner, label, value, and condition. Mood alone does not establish a counter or a completed event.

5. COMPRESS BY CONTINUITY VALUE. Older history can be denser. Recent events and the break point receive finer detail. Keep every necessary milestone; never pad a save to reach a fixed quota.

6. UPDATE THE SAVE; CARRY THE CONSEQUENCES. Previous save + new log becomes one consolidated state. Update affected facts, retain still-active older facts, and preserve consequential transition history. Silence does not settle a debt, heal an injury, or restore a place.

7. GIVE EACH FACT ONE HOME. Keep the detailed state in its own block; other blocks reference it when needed. Remove repeated wording and unsupported optional fields. A material unknown remains —, never a made-up answer.

BUILT INTO TOOL 1 & TOOL 2
═════ △ ═════
Inside the Save

These are the working memory blocks. Each keeps a different part of continuity readable. MY BOT preserves the visible state mechanics in bots created by me; ANY BOT uses the other bot’s own visible mechanics.

【SAVE BOUNDARY】
The last played line or action covered by this save, plus useful earlier coverage. Generated openings remain outside established history until played.

【WORLD CANON】
Setting: current chat-established world / faction / locale
Hard Facts: still-relevant developments and discoveries
Rules: demonstrated or explicitly established constraints

【CHARACTER PHYSICAL CANON】
Lasting injuries, marks, transformations, acquired changes, and continuing limitations established or revealed in the chat.

【SPATIAL & TEMPORAL LOCK】
The exact present: time, place, active cast, position, contact, wardrobe, current body state, and unfinished action.

【LOCATION REGISTRY — PERSISTENT ENVIRONMENT】
Place / room / route → established identity and distinctive layout → current condition → meaningful change and consequence. Add access, control, knowledge, or last-observed detail when it matters.

【CHARACTER SNAPSHOT】 — ANY BOT
Current engagement, supported mood or expression, physical tells, quirks, and voice pattern.

【INDIVIDUAL STATE VECTORS】 — MY BOT
Chat-evidenced character state, voice, and applicable visible mechanics.

【RELATIONSHIP TRAJECTORY】
Earned changes → decisive causes → current dynamic and unresolved pressure.

【SECRETS LEDGER】 / 【SECRETS — FULL LEDGER】
Concealed or revealed information, who knows or believes which version, and any established restrictions.

【RECURRING SYMBOLS / MOTIFS】
Motif → meaning earned in this RP → present continuity value.

【CALLBACK LEDGER — VERBATIM QUOTES】
Exact words → speaker and target → original function and callback value.

【OBJECTS / PROPS LEDGER】
Object → current or last-known location and condition → custodian → practical or symbolic value.

【PROMISES & DEBTS】
Who owes what to whom → pending / fulfilled / broken / modified → remaining consequence.

【NPC REGISTRY】
Established identities, latest known state, and active or off-screen consequences.

【LEXICON】
Established word or phrase → its exact meaning in this chat.

【SCENE SAVE POINT】
Context → last completed beat → unfinished dialogue or action → whose response is pending.

【CHRONOLOGICAL INVENTORY】
Necessary milestones in played order, ending at the save boundary. Recent events carry finer detail.

💉 The Double Pass — from archive to restart

Give Tool 2 both the raw log and the Tool 1 report. The report carries the consolidated state; the log supplies exact wording, tactile details, visible mechanics, and final-moment physics. Tool 2 builds the complete bootstrap from those sources.

Its Scene Start is an actual RP opening, usually 4–6 rich, atmospheric sentences: the character’s voice, sensory detail, meaningful micro-movements, and a clear opening for your response. It begins at the saved moment and leaves your speech, action, feelings, and consent to you.

✅ Before you paste the save

✓ Important characters, relationship changes, secrets, and knowledge boundaries remain intact.
✓ Callbacks, recurring motifs, promises, and meaningful objects retain their exact role.
✓ Injuries, contact, unfinished actions, and the break point agree with the last played moment.
✓ Recurring places retain their identity, layout, changes, access, and control.
✓ The new report carries still-active consequences from the previous save.
✓ A bootstrap keeps established history separate from its new Scene Start.

READ BEFORE USE
═════ △ ═════
VERSION · BUILD · CONTENT · MODELS

◷ VERSION: PC + Android universal export workflow — 22.09.2026 · v3.1

⚒ BUILD: 2 primary export workflows · 1 Legacy fallback · 4 complete prompts · MY / ANY × State Save / Bootstrap

⌬ CONTENT: Long-RP continuity · PC and Android chat export · Chat Memory · context preservation · location registry · territory state · new-chat bootstraps

⚙ MODELS: DeepSeek V4 Pro · DeepSeek V3.2 · GLM 5.3 · Claude Opus 4.7 · Kimi K3.

UPDATE · v3.1
═════ △ ═════
WHAT CHANGED · PC + ANDROID

Hey, folks — this guide used to have one annoying catch: downloading a complete chat was basically a PC-only job. Not anymore.

Android users can now install Firefox and the JanitorAI Chat Downloader, Reader & Exporter, then follow the same exported-log workflow used on desktop. Method 1 and Method 3 now work properly on both PC and Android.

The guide is now officially universal across PC and Android: ANY BOT or MY BOT, Tool 1 or Tool 2 — the same complete workflow, without the old desktop-only wall.

FURRY AUTHORS · DISCORD
There are already 6+ Janitor authors in our little Discord group chat. We help each other with art, bots, and Lorebooks, swap ideas, and just hang out.

Want to see who’s in the group? Open my profile and check the Friends section.

Russian-speaking furry authors are especially welcome. If you’re interested in what we do, say hi too. Message @000042 on Discord, send me your Janitor profile, and let’s chat.

It’s a small group chat, not a public server with random joins.

⧈ ANDROID EXPORTER OPEN

⧈ SUGGEST A BOT OPEN

⧈ LOREBOOK & SCRIPTS GUIDE OPEN

⌗ TAGSchat memory, memory guide, memory prompts, roleplay continuity, long roleplay, context preservation, state save, chat summarization, location registry, persistent world state, chat bootstrap, chat migration, chat reset, JanitorAI, JanitorAI export, Android Firefox, mobile chat export, chat downloader

GUIDE / METHOD

═════ △ ═════
Compress the wording. Preserve the state. Continue the same story.