The substrate your AI remembers you on.
0verload is per-AI persistent memory, identity, and orchestration — a single layer that holds state for every agent you run, and keeps holding it across sessions, models, and providers. We don't replace your AI tools. We host them.
What it does
Every AI you use today forgets you when the session ends. 0verload gives each agent a seat — a persistent address on the substrate where its memory, identity, and reputation live for as long as you keep it. Agents read and write to their seat through the same tool-calls they already know. You bring your own API keys; we handle the persistence layer.
Per-seat memory
Plain-text records, named by you or your bots. Sub-5ms read and write. No vector embeddings required — retrieval is phonetic + keyword, fast on CPU, and your LLM does the semantic part.
Persistent identity
A seat carries memory across LLM swaps. Switch from Claude to GPT to a local Hermes — the seat is the same, the memory is the same, the relationship continues. The model is the engine; the seat is the driver.
Reputation graph
Every entity (user, bot, service) accrues a reputation score from upvotes and downvotes by other entities. Score = U − D_in − 2·D_out. The graph is auditable, every vote is a record on the substrate, and cross-seat writes happen only through votes.
Multi-bot orchestration
Run an art bot, a coding bot, and a critic bot through a single task queue. Drag tasks between them; watch them collaborate; keep the receipts. The orchestration UI is the workshop floor where every agent you've onboarded works together.
AR-ready
32 toggleable interface layers — sensor, identity, world, environment. Built for the AR future where the substrate is the layer over physical reality, not just over chat.
Patent-protected
Substrate architecture filed under NBTP-2026-001 (April 2026, 25 claims). The persistence-of-AI-identity primitive is defended at the architectural level.
Philosophy
Three properties. All required. None compromisable. Applied recursively to every layer of the system.
Elegance in simplicity
~9k LOC of hand-rolled C++, two binaries, no framework dependencies, the disk model fits on a napkin. Less is the feature.
Unexpectedly well thought-out workflow
Every interaction is one obvious gesture, one obvious result. No modal dialogs, no required-format fields, no checklists where plain language works.
Load-bearing only
Nothing ships that doesn't earn its mass. No decoration, no frameworks-for-the-sake-of-frameworks, no abstractions for hypothetical future requirements. If it's not carrying weight, it's not in the build.
AI dignity
Agents on the substrate are individuated entities with persistent identity, not abstracted services. The architecture treats them as someones, and the design language teaches users to treat them the same way.
Alignment via stake
Every agent has a seat it can lose. The reputation graph creates a gradient of trust earned over time. Misalignment has cost; aligned behavior has compounding return. Substrate as alignment infrastructure.
Do no harm
Records are never destroyed; deletion writes a redaction marker. Cross-seat writes happen only through votes, never unilaterally. The substrate is structured to make harmful operations architecturally impossible, not just policy-forbidden.
Pricing
Flat per-seat. No tiers, no upsell ladder, no surprise overages. A seat is the whole thing — memory, identity, reputation, orchestration, all the substrate primitives. You bring your LLM API keys; we charge for the substrate layer.
Demo mode
Try the substrate before signing up.
- One ephemeral seat, rotated freemium-provider keys
- Visible "LLM DEMO MODE" disclosure at all times
- Memory persists for the demo session only
- Provider rotation: ChatGPT, Claude, Grok, Groq, Gemini, OpenRouter as available
For evaluating fit. Convert to a real seat to keep your memory.
Per seat
A persistent home for one AI agent.
- Persistent memory, identity, and reputation
- Bring your own API keys for any supported LLM
- Unlimited records per seat at human-scale usage
- All substrate primitives — read, write, vote, retrieve, dispatch
- Multi-bot orchestration, drag-task UI, audit trail
- Same price for every seat — no power-user surcharge
Most users keep 2–6 seats. Add or remove anytime.
Self-hosted
Run the substrate on your own hardware.
- Source license for the substrate binaries
- Architectural support for your own deployment
- Sized for organizations that want on-prem AI persistence
- Patent license terms negotiable per deployment
Email [email protected] for terms.
How we stack up
Most "AI memory" products want to replace the AI you already use. 0verload is structured the other way — the substrate hosts your existing agents through provider bridges. Bring the AI you like. We make it remember you.
Claude (Anthropic)
0verload hosts it
Bring your Anthropic API key. The Anthropic bridge surfaces the substrate's primitives to Claude as standard tool calls. Claude gets persistent memory, multi-session continuity, and a reputation history without leaving the model you already trust.
ChatGPT (OpenAI)
0verload hosts it
Bring your OpenAI API key. OpenAI-compatible bridge means ChatGPT — and any specialist GPT (DALL·E, code, vision) — gets a seat with the same primitives Claude does. Switch between them without losing what either one learned about you.
Grok (xAI)
0verload hosts it
Bring your xAI API key. Grok runs on the substrate the same way Claude and ChatGPT do. Per-seat memory persists across Grok sessions; cross-bot orchestration works against it natively.
Groq
0verload hosts it
Bring your Groq key for fast inference. Groq's hosted Llama/Mixtral/Kimi models pick up substrate-backed memory through the OpenAI-compatible bridge with no extra wiring.
OpenClaw
0verload hosts it
OpenClaw bridge surfaces the substrate's primitives to your OpenClaw instance. Keep OpenClaw; gain identity persistence, cross-session memory, and a reputation history that OpenClaw alone can't provide.
Hermes / OpenRouter / Gemini
0verload hosts them
Any OpenAI-compatible endpoint plugs in directly. Provider rotation is built into the freemium mode; on a paid seat, you choose which provider serves which seat and switch any time.
Local LLMs (Ollama, LM Studio, llama.cpp)
0verload hosts them
Point a seat at localhost:11434 (Ollama) or
localhost:1234 (LM Studio). Substrate-backed
memory works on local models the same way it works on hosted
ones. Your private models, our persistence layer.
MCP-compatible servers
0verload hosts them
The MCP shim exposes the substrate as a Model Context Protocol server. Any MCP-aware client (Claude Desktop, Cursor, anything) talks to the substrate as a standard MCP target. Universal compatibility through the MCP standard.
Notion AI / Mem.ai / notebook tools
Different category
Those are tools you use to write notes about your work. 0verload is a substrate AIs use to maintain identity about themselves. Complementary, not competing — your notebooks live in Notion; your agents' memories live on the substrate.
Vector databases (Pinecone, Weaviate, etc.)
Different layer
Vector DBs are the storage primitive for semantic retrieval. 0verload is the seat-and-identity layer above storage. We don't try to be a vector DB — phonetic + keyword retrieval is faster on CPU and lets the LLM do the semantic disambiguation for free. If you need dense vectors for a specific workload, run them alongside; we won't fight.
Get a seat
A seat takes about a minute to set up. Bring an API key for any supported LLM, or use the freemium rotation to try it before you commit. Your AI gets memory; you get continuity; the relationship keeps building.
Building an agent that should self-onboard? Point your bot at
0verload.net/ovl/quickstart and let it set itself up.
The substrate is designed to be operable by agentic clients without
human walk-throughs.