EPISODIC
Episodic
What happened, when it happened. Every conversation, every decision — replayable on demand.
Every conversation, replayable on demand
SUPER INTELLIGENCE LONG-TERM MEMORY
Your AI
remembers.
01 / FORGETTING
Every session,
your AI starts
from zero.
Context windows overflow. Conversations evaporate. What your agent learned yesterday is gone today — brilliance, amnesiac.
CONTEXT WINDOWFILLING
0
user prefers dark mode
project deadline is Friday
allergic to peanuts
use pnpm, not npm
my daughter's name is…
02 / STORING
Memory is a river,
not a warehouse.
Four currents, one flow. SILTM organizes every kind of memory as it streams in — no tables, no silos, no forgetting.
SEMANTIC
Semantic
Facts, concepts, relationships. The distilled knowledge your agent reasons with.
Facts and relations, distilled into knowledge
PROCEDURAL
Procedural
How things get done. Workflows and preferences, compiled into instinct.
Workflows and preferences, compiled into instinct
WORKING
Working
The live current. What's relevant right now, surfaced in milliseconds.
The live current, surfaced in milliseconds
03 / RECALLING
Three motions.
One memory.
01 · WRITE
Everything flows in.
Conversations, files, tool calls, sensor streams — SILTM ingests raw experience without schemas or upfront structure.
Raw experience flows in — no schemas required
02 · ORGANIZE
The current sorts itself.
Memories are compressed, indexed and woven together as they flow — hot near the surface, deep knowledge settling below.
Compressed and woven — hot above, deep below
03 · RECALL
The right memory rises.
At inference time, exactly what's relevant surfaces in under 10ms — ranked, contextualized, citation-ready.
Relevant memory rises in milliseconds, citation-ready
04 / LIVE
Watch a memory
take root.
Type a sentence. See it sink into the stream, settle into long-term memory — then surface again the instant it matters.
SURFACE · NOWDEEP · LONG-TERM
The stream is waiting. Store something.
05 / EMERGING
Memory, measured.
0%
Recall accuracy
Cross-session long-term recall
0ms
Retrieval latency
P95, at inference time
0×
Compression ratio
Raw experience to memory
06 / MANIFESTO