Download Now
v1.0.0 — Now Live

Your AI
Forgets You.
We Fix That.

Close a chat, open a new one, and it's like you're a stranger again. UPtrim sits between your chat app and your LLM and gives it a real memory. Names, preferences, old conversations — all retained automatically.

Free on GitHub No account required Runs 100% local Drop-in for any OpenAI-compatible app
How it works
Your Chat App UPtrim Your AI
Persistent memory Context trimming User isolation File uploads
0 Context Lost
Private Memory Per User
2 Min Setup Time
Free Developer Key on GitHub
Compatible apps
Open WebUI SillyTavern llama.cpp Ollama Any OpenAI-compatible app
the ghost remembers how to play
GPU backends
NVIDIA CUDA AMD ROCm Apple MLX

What It Does

What changes when UPtrim sits between your chat app and your LLM.

✂️

Context Trimming

Long chats degrade as older messages fall out of the context window. UPtrim compresses earlier turns and keeps the important parts visible.

Click to learn more
🔒

Multi-User, Isolated

Multiple users, one proxy. Each gets isolated memory and files — nothing crosses over.

Click to learn more
🧠

Persistent Memory

NLP extracts facts from your conversations — name, preferences, projects — and stores them in SQLite. They show up again when relevant.

Click to learn more
📄

File Uploads

Upload PDFs, code, spreadsheets, docs. The LLM pulls relevant chunks mid-conversation. 33+ formats supported.

Click to learn more
📊

Web Dashboard

View and edit memories, manage users, change settings. Real-time stats at localhost:9099.

Click to learn more
🔗

Drop-In Compatible

Change your API URL to localhost:9099. Works with Open WebUI, SillyTavern, and anything OpenAI-compatible.

Click to learn more

Setup Takes 2 Minutes

Download, run, change one URL.

1

Download

Grab the release for your OS from GitHub. Run it.

2

Redirect

Change your chat app's API URL to localhost:9099. UPtrim forwards everything to your LLM.

3

Done

Chat normally. Memory builds in the background. Open the dashboard to see what it knows.

The Dashboard

Memories, users, stats, settings. All at localhost:9099.

localhost:9099/dashboard
U
UPtrim
v1.0.0
📊 Overview
🧠 Memories
👥 Users
📁 Files
🤖 Agent
⚙️ Settings
🌐 Tunnels
📝 Logs
Online
Overview
System health and recent activity
Ollama ✓ NLP ✓
247
Memories
5
Users
12
Files
1.2k
Messages
Active Users
A
Alice
42 memories
B
Bob
28 memories
C
Carol
67 memories
Recent Memories
Name is Alex, prefers Python identity
Works at Acme Corp, backend team work
Prefers concise technical answers comm
Allergic to shellfish health
Daughter Emma starts school in Sep family
Click the sidebar to explore. This runs at localhost:9099/dashboard.
psst... type “snake” anywhere
🔑

Free Key on GitHub.

v1.0.0 ships with a developer key in the README. No sign-up required, no expiration.

🔑

Free Developer Key

Paste it when UPtrim starts. Memory, multi-user, file uploads, and the dashboard all work out of the box.

Get the Key on GitHub Windows • macOS • Linux
Persistent memory across chats
Context trimming for long sessions
Up to 5 isolated users
File uploads (33+ formats)
Full web dashboard
NLP extraction with spaCy
👻

Paid Tiers

The free key gets you started. Paid licenses add background agents and higher limits.

Standard+
👻

Ghost Agent

Micro-agents that fire on every message. They fetch URLs you paste, run web searches, and pull from deep memory. Results land in the system prompt before the LLM responds.

Paid License
🌿

Memory Gardener

Runs when nobody's chatting. Merges near-duplicates, resolves contradictions, and builds per-user profile summaries. You don't have to manage it.

Coming Soon
🕸️

Knowledge Graph

Memories stored as connected nodes instead of flat rows. The LLM sees that Sarah leads Project Atlas, not just that both exist independently.

Coming Soon
🎭

Persona Engine

Learns how each user likes to be talked to. One person gets terse technical answers, another gets friendly walkthroughs. Adapts per-person automatically.

Stop Losing Context.

Free developer key on GitHub. Paid licenses add Ghost Agent and 1 GB uploads.

ever tried racing through memory? or dodging asteroids made of bugs?

Continue reading See How It Works