INGESTION / INDEXING / EVIDENCE RETRIEVAL

Search over source observations.

Atheros Search combines keyword and vector retrieval over configured proxy and wireless observations. RAG builders and operators can inspect source records, embedding jobs, and ingestion state in one workbench.

Follow ingestion

Trace configured sources, processing, and freshness.

Inspect indexing readiness

Separate lexical evidence from available semantic coverage.

Retrieve supporting evidence

Review ranked records and explanations. Answer generation is deferred.

Atheros Search / illustrative site review

Synthetic sample

This sample illustrates a site-to-indicator-to-observation path. The current production console does not show this complete site overview end to end.

2. Review an indicator for analyst review

Suspected rogue access point

North Campus / sample / Sensor 02 / sample

SSID lookalike heuristic / sample

3. Review why the indicator was raised

A sample beacon advertises an SSID that resembles a configured network name. The sensor can raise this configured heuristic as an indicator for analyst review. The match alone does not prove a rogue access point.

4. Inspect the supporting observation

Observed access point 07 / Beacon observation / sample

Channel 6 / 09:42 UTC / sample time

Sensor 02 recorded a beacon from observed access point 07 on channel 6 at the sample time. This synthetic relationship does not prove an active connection or physical device identity.

Sensor placement, configured channel coverage, observation time, and MAC randomization limit what can be seen. An observed Wi-Fi identifier does not confirm a current connection or physical device identity.

Ingestion and retrieval

The authenticated workbench includes retrieval, sources, ingestion, and processing. This synthetic sample follows a site, an indicator, and its observations. The current production console does not yet show this complete site overview end to end. Generated answers and new connectors remain future work.

  1. 01 / Observe and ingest

    Observe and ingest

    Input
    Configured proxy and wireless observations
    Processing
    Publish events. Octopus records evidence and deduplicates repeated delivery
    Output
    Durable ingestion evidence within configured source coverage
  2. 02 / Prepare and index

    Prepare and index

    Input
    Ingested observations and maintained projections
    Processing
    Octopus prepares documents and jobs. Fenced Search workers create embeddings
    Output
    Searchable documents and available vector coverage
  3. 03 / Retrieve and inspect

    Retrieve and inspect

    Input
    A query and supported evidence filters
    Processing
    Rank term and vector matches. Inspect supporting records and explanations
    Output
    Evidence for human review or a downstream RAG workflow

Working with Search

RAG builders and operators who need retrievable evidence, visible processing gaps, and source context for human review.

FOR RAG BUILDERS

Retrieve records and their source context.

  • Search dense, sparse, or hybrid records through existing HTTP and gRPC interfaces, with supported filters and ranking explanations.
  • Distinguish available lexical evidence from semantic coverage. Missing vectors, failed embedding jobs, and unknown measurements do not establish indexing readiness.

FOR OPERATORS AND REVIEWERS

Check ingestion and indexing failures.

  • Inspect stages, backlog, failures, workers, and measurement freshness alongside retrieval, using the existing login and role permissions.
  • Separate indexed proxy evidence from operator-only live aggregate telemetry. Operational readings are global unless explicitly scoped. Missing data stays unknown.

Implementation details

Listen on configured Wi-Fi channels

Atheros Sensor uses monitor-mode capture. It listens on configured channels rather than joining access points, then publishes audit records to the configured backend.

Review configured wireless indicators

Sensor heuristics can flag suspected rogue access points, deauthentication floods, signal anomalies, attack sequences, and PMF-related patterns for analyst review.

Inspect supporting observations

Review wireless audit records with site, sensor, channel, and observed-time context. Retention details depend on deployment.

Treat identifiers carefully

Inventory shows observed Wi-Fi identifiers. MAC randomization and incomplete sensor coverage mean identifiers are not a count of confirmed physical devices.

See processing progress

The ingestion and sources views expose allowlisted processing and coverage measurements. Inspect embedding jobs, worker heartbeats, failures, and timestamps without treating unavailable data as zero.

COVERAGE AND IDENTITY

Sensor placement, configured channel coverage, observation time, and MAC randomization limit what can be seen. An observed Wi-Fi identifier does not confirm a current connection or physical device identity.

The sensor listens on configured channels and publishes audit records to the configured backend. Deployment requires monitor-mode Wi-Fi hardware. Audit fields and retention depend on deployment. No retention duration is promised here.

Terms

Dense search
Finds records with similar meaning using numerical representations called embeddings.
Sparse search
Finds records by matching terms.
Hybrid search
Combines meaning and term matches to rank results.
Wireless indicator
A configured sensor heuristic that points to a wireless pattern for an analyst to review. It does not prove a threat.
Observed Wi-Fi identifier
An identifier seen in captured wireless traffic. It may not map one-to-one to a physical device.
Observed relationship
A link supported by recorded observations. It does not prove a current connection or confirmed identity.
ETL
Extract, transform, load: the processing steps that prepare data for use.

Technical guides

Worked examples for SQL migrations, wireless investigations, and hybrid search.

Hybrid search over network observations

Compare term, vector, and hybrid retrieval on the same synthetic observations. Work through weighted reciprocal-rank fusion and interpret explain data and sparse fallback.

CONTACT

Questions about a project?

Email me with the project, environment, and constraints you want to discuss. We can arrange a walkthrough or technical call by email.

Email Rafael

Opens your email app, or email rafael@rclabs.uk. Start by email. We can arrange a technical call together.