Deep Learning Sandbox  ·  sloman.ai ● RELEASING MAR 12

No Cloud.
No Compromise.
No Permission Required.

SandboxDL is a local-first AI companion built for people who want to own what they build. Custom personalities. Persistent memory that lives on your machine. Full control over the model, the data, and the relationship — with zero bytes leaving your device.

Releasing March 12 See Features Beta release March 12
sandboxdl.exe
›Initializing SandboxDL...
✓ Core identity loaded
✓ Memory engine online
✓ Semantic retrieval ready (384-dim)
  Model: qwen/qwen2.5-vl-32b · OpenRouter ● connected
›session.start("Murphy")
Murphy: Ready. What are we working on?
›█
100%Local Storage
∞Persona Depth
BYOKBring Your Own Key
βBeta Mar 12
Capabilities

Built Different

🧠

Persistent Semantic Memory

Your AI doesn't forget. Conversations are embedded and stored locally using high-dimensional vector search — surfacing relevant memories automatically, exactly when they matter.

ACTIVE
✏️

Live Identity Control

Reshape your AI's core personality, system behavior, and long-term memory in real time through native dialog panels — no config files, no restarts, no friction.

ACTIVE
🔑

Bring Your Own API Key

Connect any OpenRouter-compatible model. Swap between frontier models or run lean — your key, your costs, your choices.

ACTIVE
🗂️

Intelligent Memory Archiving

Active memory auto-archives aging conversations, then prunes the archive into distilled long-term summaries — keeping your AI sharp without bloating context.

ACTIVE
🃏

Character Card Exports

Export fully-formed AI personas as portable JSON character cards — shareable, importable, and structured for use across compatible platforms.

FULL RELEASE
🧬

LoRA-Ready Training Exports

Export conversation histories as structured fine-tuning datasets — formatted and ready for low-rank adapter training pipelines.

FULL RELEASE
Memory Architecture

Memory That Actually Works

Every mainstream AI chat is stateless by design — each session starts from zero. SandboxDL is architected around the opposite assumption: that memory should compound, context should carry forward, and your AI should get sharper the longer you work with it. Three layers, fully local, always on.

01
Active Context

Recent exchanges held in working memory — fast, immediate, always current.

02
Archive

Older exchanges automatically compressed into dense narrative summaries.

03
Long-Term Vector Store

Semantically searchable memories retrieved on demand — always accessible, never forgotten.

Roadmap

What's Coming

Beta Release
March 12
● UPCOMING
  • Full Chat Interface — polished, responsive desktop UI
  • Live Identity Editor — real-time persona shaping
  • Semantic Memory Engine — multi-tier vector memory, always on
  • Multi-Model Support — swap models without restarting
  • Memory Prune & Archive — smart compression of long history
Full Release
March 14
PLANNED
  • Character Card Exports — portable JSON persona snapshots
  • LoRA-Ready Dataset Exports — structured fine-tuning output
  • Persona Versioning — snapshot and restore any identity state
  • Session Analytics — memory stats, model usage, conversation depth
  • Much More — the sandbox is just getting started

Stop Renting Intelligence.
Start Owning It.

Beta drops March 12. Full release March 14. Free to run — bring your own API key and keep everything that happens between you and your AI exactly where it belongs: on your machine.

Releasing March 12 ← Back to sloman.ai