Grid Theory · Research
Grid, the co-residence language.
An append-only, capability-projected cell substrate where the agent's cognition and the system's state are the same artifact. Continuous presence as a structural property, not a feature we'd have to ship.
A single grammar that residents (humans and Ai) speak. One executor for every transport. Capability is projection, not access control. Denial is silence.
A grid expression, same shape from CLI, REPL, HTTP, or a resident's attention cell.
# Every cell that promised an action before today, as a one-line digest:study: type=interaction where: slot:promised < today as: line# What did Cameron Silva look like the day before she churned?study: history-of: @/crm/contacts/cameron-silva as: line# Stream every inbound interaction as it lands:watch study: type=interaction where: slot:direction = inboundThree pillars
What makes Grid Grid
Language
One grammar, closed under composition
Six axes (select, project, filter, mutate, observe, compose). Every operation reduces to them. CLI, REPL, HTTP, WebSocket, and resident attention cells all bottom out in one evaluator. There is no second path.
Read more
Capability
Deny-by-invisibility, not access control
A capability cell IS a grid expression. Two identities with different capabilities inhabit different grids even when the same file is on disk. Denied cells return empty results, never "permission denied." Errors must not leak existence.
Read more
Residence
The agent's cognition and the system's state are the same artifact
Residents are long-running tick loops bound to a grid: humans at a CLI, programs in a loop, or models with capability cells, all writing into the same chain under one grammar.
Read more
When we bring Grid (and when we don't)
Grid Theory is a company and a literal theory: a company that ships software, and a body of locked claims about what software should be when the agent and the substrate are the same artifact. Grid is the substrate we built. We use it for AI-native platform work, where residents matter, where memory has to survive sessions, where the read model and the cognition graph need to be the same object. For other workloads (CRUD apps, marketing sites, SQL workloads, JSON APIs, mobile) we use the right standard tool. The dual-stack honesty is the discipline.
Status
- Kernel: Native Rust. Append-only chain integrity, append log, capability projection.
- CLI: Native single-binary tool. Expressions, REPL, watch, history, encryption.
- Console: Native iced GUI. Cell viewer, write form, history, keychain, filter bar, projection chips, crypto modals (lock, unlock, seal, unseal with recovery code). Pre-release.
- Resident: Native single-binary tool. Three flows: mirror (sensitivity-filtered cross-grid replication), digest (one-line summaries), auto-seal (threshold-triggered grid sealing with keychain anchors).
- Capability model: Spatial, four-tier sensitivity lattice.
- Public download: Not yet available.