Semantic state layer for reproducible AI
Saving, replaying, transferring and auditing the state of AI agents in production.
New to SEMQ? Start with the overview →
Use Cases
Reproducibility
Deterministic encodings make AI pipeline outputs stable across hardware and time.
Portability
Hardware-agnostic encoding, so you can move workloads freely without state migration pain.
Versioning
Diff model outputs across versions like source code and catch semantic drift early.
Observability
Structured angle sequences make AI state auditable and loggable for the first time.
Caching
Stable angle keys enable exact-match and approximate semantic caches at scale.
Compression
16× smaller state footprint than traditional PQ.
The numbers speak for themselves
Measured on public datasets. MTEB, BEIR and OMB v1.
vs −36pp for PQ
Accuracy loss on banking77 (77-class). Same bit budget, 1,200× less damage.
over full context
On LongMemEval-S. SEMQ retrieval beats stuffing the entire transcript into a 1M-context model.
9,511 queries
Bit-identical results across 1,000+ concurrent processes. Cross-architecture. No exceptions.
Embedded on GPU A
[0.18291, −0.04213, 0.99187, …]
Same text, embedded on GPU B
[0.18288, −0.04209, 0.99191, …]
≠ Different bytes downstream — different cache keys, different diffs, different results.
The Problem
Modern embeddings are architecturally broken for production AI.
They are hardware-dependent, non-reproducible, impossible to diff, and expensive to store at scale. Every float changes when you switch hardware. Two embeddings of the same text on different machines produce different bytes.
As systems grow larger, more distributed, and more persistent, these limitations compound. Memory becomes unstable. Routing becomes probabilistic. Storage and transmission become increasingly costly. The field has been working around this with heuristics, versioned indexes, and expensive re-indexing cycles.
θ12 · θ34 · θ56 · …
The Approach
What matters for finding relationships between vectors is the topology of the vector space.
SEMQ encodes pairs of vector components as angular coordinates. Each angle represents the local relationship within the embedding. Taken together, the angle sequence is an address in semantic space that is independent of floating-point precision or hardware rounding.
This encoding preserves pairwise similarity within a configurable error bound. As the number of angle dimensions increases, the bound tightens.
Converts vectors into symbolic angle codes
[0.183, −0.042, 0.991, …]
SEMQ code
θ12 · θ34 ·
θ56 · θ78 ·
…
Architecture
A new primitive for AI systems.
Symbolic angles integrate naturally into existing infrastructure. They are small enough to store in a database column, stable enough to use as cache keys, and structured enough to diff across model versions.
SEMQ provides the semantic state layer that slots into your existing AI stack, either as a portable .semq file or via the SEMQ MCP server.
Be first to ship with SEMQ.
The SDK lands soon. Join the waitlist to get early access and updates as we ship.
We'll only email you with SDK updates.
