SOFTWARE

Permission-aware enterprise search and actions for a software company

Employees needed one place to ask questions across chat, email, documents and HR policies, and to act on the answer. Wexa built an assistant that answers only from what each person is allowed to open, cites its sources, and performs actions after confirmation and approval.

Download this case as PDFBook a demo
tokens vs sending raw data to the model (up to −82%)
correct answers vs 8 / 20 without Wexa
permission leaks across 9 probes
hit rate in the top 3 results
CLIENT
Enterprise software company (in-house deployment across chat, email, drive and HR policies)
REGION
US / India
WHEN
2026
ENGAGEMENT
Solution build
WEXA COMPONENTS
Wexa · connectors · automatic ontology discovery · Wexa memory layer · approvals

01Case 09

The challenge

  • Knowledge scattered across chat, email, documents and policies
  • Search tools that ignore who may see what
  • Answers that cannot be acted on safely

What Wexa built

  • Ingestion of files and messages with automatic ontology induction into the graph
  • Hybrid retrieval: keyword, graph and knowledge base, run as the signed-in user
  • Tool-calling agent that plans, cites and loops until the task is done
  • Writes confirmed in chat, approved by a different person and executed exactly once; every step audited

02Case 09

How it runs on Wexa

FigureHow this use case runs on Wexa: client systems connected, ontology discovered, context graph in the Wexa memory layer, agents and process flows, and governance. Highlighted components are central to this use case.

03Case 09

Results

MeasureBefore / without WexaWith Wexa

Hours a week each employee spends looking for information (30 people)

2.5 hours

1 hour

Annual cost of the process

$144K

$58K

Time to implement the AI solution

3 months

1 week

AI build cost (one-time)

$96K

$4K

AI operating cost per year (models, infrastructure, upkeep, platform)

$65K

$29K

Total AI cost in year 1

$161K

$33K

With Wexa

net annual value to the client after Wexa's own operating cost

faster to implement

lower AI operating cost

less AI spend in year 1

  • Started as the enterprise-agent track of the October 2026 hackathon
  • API stress test: 47 routes, worst p95 under 380 ms at 12 concurrent sessions, no server errors

More case studies

All 43 cases
08Solution build

Consumer packaged goods

On-shelf availability control tower · CPG

−72%

tokens per case vs the same agents without Wexa (59,269 → 16,559)

72–81% fewer tokens than the same agents without Wexa; 58% lower cost; actions written only after approval

Read the case
11Solution build

IT service management

AI service desk · enterprise IT

5.5 s

self-service answer citing past incidents

Answers in 5.5 s, escalations in 4.5 s; go-live in 9.4 s; built entirely on Wexa

Read the case
13Solution build

Energy · wind

Wind-blade failure containment · turbine OEM

6 vs 24

inspections needed (graph cohort vs PLM-only)

6 inspections instead of 24 with 100% recall in validation; no defective blade left in service

Read the case
08 On-shelf availability control tower · CPG10 CodeMesh · context for coding agents

Want results like these?

Book a 30-minute call and we will map your workflow to this case.

Book a demo