01
Speed without continuity becomes drift.
Documents, chats, tickets, and dashboards are useful, but they rarely preserve the operational meaning behind a decision.
Foundation story
The graph remembers. The model reasons. You ratify.
Agentic systems need more than prompts and documents. They need a substrate where meaning persists, change stays legible, and human judgment compounds.
coherence, not compute
typed graph as substrate
controlled change near action
Scroll
01 / The arc
Capability is abundant. Coherence is scarce.
Vellis starts from the failure mode that shows up once teams add more models, tools, agents, and automations: everyone can move faster, but the system can stop agreeing with itself.
02
Text alone cannot carry the system.
Agents can read text, but execution needs typed things: who can change what, which link means dependency, what counts as approved.
03
The graph is a theory about the work.
A graph is not just storage. It is a living model of what the organization believes is true enough to act on.
04
Control belongs where execution happens.
Vellis keeps knowledge change explicit and schema-validated. The governed production layer adds policy, approvals, and audit controls when that knowledge becomes operational infrastructure.
02 / The operating loop
Work gets more useful when it leaves reusable context behind.
Each useful run should leave typed context the next agent can use. Vellis captures structure, agents work against it, humans review the important changes, and the graph becomes a stronger substrate for the next action.
01
Capture
Store typed nodes, links, provenance, and decisions.
03 / The felt need
Agents need explicit knowledge, not another document pile.
The failure mode is familiar: chats age out, documents drift, and every new agent session has to reconstruct what the team already learned.
Where teams feel it
The work is moving, but the shared context does not keep up.
Agents forget useful context between
sessions.
Docs multiply faster than people can keep
them coherent.
Decisions get trapped in transcripts,
tickets, and stale notes.
What the graph changes
The system gets a durable model of the work, not just more text to search.
Meaning lives in typed nodes and links.
Provenance and change history stay attached
to the knowledge they describe.
Future runs start from current structure,
not reconstructed summaries.
04 / Published thinking
Perspectives carry the ongoing conversation.
Essays and field notes live in one place, so the library can grow without turning this page into a blog index.
As the workflow matures, graph-drafted pieces can carry visible provenance and human review status. Until that workflow is real, published work should read as authored Volant Labs thinking.
Run Vellis with the model in view.
Start with the open engine, then keep what matters: shared meaning, recoverable state, and human ownership of the graph you build.