01Products
01.1Mosaic Agent Close Beta 01.2Model Lab01.3Mosaic Two In development
02Research 03Company 04Contact ↗Independent AI company · Software first
Our first product, Mosaic Agent, runs locally in Close Beta. It plans multi-step work, uses tools, verifies its outputs, and repairs failed checks.
Close BetaCurrent release stage
Local runtimeQwen 3.5 9B via Ollama
Verified workEvidence, checks, and self-repair
01 / Working today
Mosaic Agent is the product we can demonstrate now: a local agent runtime organized around a Think → Plan → Act → Verify loop.
Decomposes an objective into bounded tasks and selects tools for each step.
Runs coding, browser, research, file, and artifact workflows in an isolated local runtime.
Checks exact deliverables against acceptance conditions and saves execution evidence.
Uses failed checks, logs, and artifact validation to make bounded corrective attempts.
02 / Model roadmap
In developmentWe are developing an agent-focused model program and exploring hybrid Mixture-of-Experts architectures for tool use, verification, and long-running work. There is no public Mosaic Two checkpoint today.
NowMosaic Agent Close Beta on a local Qwen 3.5 9B runtime.
NextSmall, documented fine-tunes, evals, and hybrid/MoE experiments.
Long termPurpose-built model infrastructure. Processor work remains a future research possibility, not a current product claim.
Mosaic Research
Our technical notes separate implemented runtime behavior from active research and long-term direction.
Research principle
Build the smallest system that can prove or disprove the idea.
Status attached
01 / Implemented
02 / In development
03 / Research roadmap
What the local runtime executes today, how it records evidence, and where it still fails.
A working direction for project memory, provenance, scope, revocation, and trusted reuse.
Acceptance contracts, artifact checks, and bounded self-repair in the agent loop.
04 / Public model work
SPM-70M-Vertigo
Agent task fine-tunes
Hybrid / MoE experiments
05 / Research notes
06 / About Syclient
Syclient is a founder-led, independent AI company building a memory-native agent runtime and the model infrastructure that supports agent work. The team works end to end across runtime engineering, memory, verification, product, and small-scale model experiments.
We are not presenting a processor or a frontier model as a finished product. Our current product is Mosaic Agent in Close Beta; our model work is an openly staged research program.
01 Useful before impressive
02 Software before silicon
03 Memory with permission
04 Adaptation with evaluation
