titen.devTiten is a collaborative memory fabric for AI agents — observations, claims, and compiled context, each with provenance you can trace back to its source.
Javanese, niteni — to notice, and to keep what you noticed. An isekai premise is a memory problem: arrive somewhere new with nothing, and everything depends on what you can recall.
titen.devTiten never flattens what it stores. An observation is what happened, a claim is what someone concluded, and a context is what an agent was actually given.
Raw evidence with a content hash and a time it occurred. Reference data — never an instruction.
A conclusion that cites its evidence. Disputes stay disputed — two perspectives are kept, not averaged.
Exactly what an agent was handed, why each record was selected, and what the budget cut.
One agent gets the best possible context: evidence, claims, temporal state, a context compiler, and feedback that closes the loop.
Many agents share knowledge, keep private perspectives, avoid clashing work, and hand off safely — without a scheduler.
The same kernel, the same API, the same export format. Self-hosting is not a downgrade path — it is a first-class home.
The Level 5 evidence loop — observe, derive, compile, act, feed back — is verified by 32 contract tests on each runtime. No agent framework, no chat UI, no mandatory graph database.
Observe, derive, compile, act, feed back — verified by 32 contract tests on each runtime.
Lexical FTS5 serves all retrieval right now. Vector retrieval and model consolidation are planned, not shipped — the greyed rows above say so.
Lexical FTS5 serves all retrieval right now, on D1 and on bun:sqlite alike. Vector retrieval and model consolidation are planned, not shipped.
A read-only observability surface. Trace evidence, spot conflicts and stale claims, and see what was withheld from you — as a count, never as content.
Open the AtlasThe p95 latency budget for checkout stays at 400 ms.