NovusLens · system mechanics
From company records to evidence you can use.
Select a flow, or inspect a component. Arrows connect services. Mycelium carries structural connections; Docker hosts services. The dotted wiki link means stored in the database.
Docker hosts the services. Mycelium provides their application mesh and structural event log. Postgres stores the app’s records. Source files and some acquired evidence enter it directly. Mapped structure also travels through Mycelium and projection. The internal wiki stores knowledge in the same database. Model calls are a separate, optional computation path.
Company files follow different paths according to their contents.
Organisation charts create documented structure through Mycelium. Policies and runbooks enter the internal wiki in Postgres, with source citations. Personnel exports link hashed identities to teams in Postgres. These sealed records stay outside the structural log and the wiki. The web server reads permitted maps from Postgres. Knowledge library operations go through the reasoning service.
Keep the source export and publish its mapped structure separately.
The importer reads supplied cloud inventories, Terraform state and datacentre exports. It stores the original source in Postgres. Domain tags and labels join documented cells and dependencies to the organisation map. Those structural records travel through Mycelium and projection. This is file ingestion, not live cloud discovery or measured telemetry.
Read operational sources and record measured observations.
Acquirers read operational APIs or supplied network-flow exports. Score and flow acquisition retain supporting evidence and calibrations in Postgres. Compact readings go through Mycelium. A separate corrector records current observations and differences in the structural log. Projection builds the readable view. These calculations do not require a language model.
Agent declarations use connector imports, rather than the signed-event outbox.
AgentFacts imports and enrolled boundary feeds describe agent authority and reported behaviour. Connectors retain source artifacts in Postgres and commit mapped declarations or findings through Mycelium. Projection updates the map. This path is separate from the signed activity-batch API. NovusLens reads the customer’s estate. It does not operate its agents.
Store the signed batch before acknowledging it. Publish references afterwards.
The reasoning service checks source enrolment and signed records at its HTTP intake. It commits the batch, source evidence and delivery record together in Postgres before acknowledging receipt. The customer projection worker reads pending batches and publishes references through Mycelium, then builds queryable records. A receipt confirms storage, not independent assurance or completed projection.
People confirm structural changes. Application code records the result.
The web app sends questions and actions to reasoning. Captured structural proposals wait for human confirmation. Reasoning records changes, resolved findings and assessments through Mycelium. Projection updates Postgres. Some operations refresh it immediately. The web server reads permitted maps and findings from the database.
Retrieve permitted records and call a model when the task needs interpretation.
The web app identifies the signed-in viewer for each request. Reasoning retrieves permitted graph records and source-linked wiki entries from Postgres. A configured model can help interpret context and compose an answer. It returns output to application code, which supplies citations and handles accepted changes. Measurements come from evidence and calculations.
Conceptual responsibilities, grounded in the Docker demo. NovusLens maps, surfaces and advises. People decide what changes to make.