Skip to main content

Hands-on agent demos. No sign-up, no sales call. Try Them Now

Book a Demo

IT · Application Development · All Industries

You Can't Modernize What You Don't Understand.

The people who understand your most critical legacy applications are retiring, taking decades of undocumented knowledge out the door with them. Squirro + RMJ2 captures scattered knowledge from source code, developer chats, tickets, and runtime behavior, structures it into a queryable knowledge graph, and turns it into the knowledge foundation needed to drive the modernization of legacy software applications.

Reduce Risk Deploy in 6–8 Weeks Medium Effort
Book a Discovery Call

Trusted by Industry Leaders

library of congress OCBC Bank Zwickroell Security Benefit Henkel Siemens Bertelsmann pavlov media long beach container terminal Candriam IDB Invest Bühler logo swissrockets Customer Logo

Measurable Impact

Colleagues working through requests together
98%+ answer accuracy, grounded in your actual application knowledge

Knowledge preserved permanently, independent of the people who held it.

What It Does

You Can't Modernize What You Can't Understand. Knowledge Is Your Foundation.

Critical legacy applications rarely fail because the code is bad. They become a liability because the people who understand them are retiring, the documentation is absent or untrusted, and the knowledge lives scattered across the organization, including in legacy source code, developer chats, ticket threads, runtime logs, and the heads of a shrinking group of senior staff. Squirro + RMJ2 uses agentic RAG grounded by a knowledge graph to answer questions across your source code, chats, tickets, and runtime behavior, retrieving cited answers your teams can rely on.

  • source code, chats, tickets, and runtime behavior ingested and structured into one queryable model
  • dynamic SQL, execution traces, and runtime logs captured, not just static code
  • developers ask in natural language and get cited answers traceable to the actual source artifact
  • answers come from your knowledge graph, not a probabilistic guess; 98%+ accuracy versus 55–70% for generic AI bots
  • permissioned access, sensitive content classification, full audit trail; source code never sent to a public LLM in raw form
Application Knowledge Assistant Ask about any program, job, screen or ticket · every answer cited to the source artifact Northline · Claims platform
Quality Manager – new to the program, no COBOL expertise An emergency fix has to be made to an online COBOL program. What is the impact, and what's the risk?
Resolved the program from change request CHG-2026-0311 · PAYCLM01, online (CICS)2.1s
Graph traversal — calls · includes · displays · 4 modules, 11 subprograms, 6 copybooks, 3 screen formats, 2 business rules3.4s
Runtime behavior — BMP schedule, fallback and SLA from job logs and monitoring config2.8s
Ticket search — 2 open records linked to PAYCLM01, checked against the change scope1.9s
Change documentation and test-plan template retrieved · impacted paths mapped3.2s
Planning context — business-owner calendar and change-freeze window1.6s
Squirro + RMJ2

Analysed the full change scope and its dependencies: 4 modules, 11 subprograms, 6 copybooks, 3 associated 3270 screen formats, and 2 business rules.1

The program runs 10–15 times daily between 22:00 and 23:00 as a BMP job, with a documented fallback restoring service within 1–2 minutes. SLA is 98% availability, currently measured at 99.5%. Business criticality: moderate.23

The workaround is a manual database update carrying a risk of data inconsistency. Two open tickets relate to this program, neither affected by the change.4

A draft test plan covering the impacted paths is attached with the relevant change documentation. The original developer has left the organization – knowledge and dependencies retrieved from the graph.5

Note for planning: the business owner is on leave for two weeks, and a change freeze begins two weeks after that.6

Draft test plan — PAYCLM01 emergency fix6 impacted paths · generated from the graph Change documentation — CHG-2026-0311linked record
1PAYCLM01.cbl · 11 subprograms · 6 copybooks3 BMS maps · 2 rule documents · graph traversal 2BMP job PAYNITE · runtime logs10–15 runs daily · fallback runbook RB-114 3Service-level config · CLAIMS-ONLINESLA 98% · measured 99.5% 4INC-48213 · INC-48377ServiceNow · open · outside the change scope 5CHG-2026-0311 · test-plan templateSharePoint · draft attached 6Business-owner calendar · change calendarleave 2 weeks · freeze window after
Answered in 17 s · source code, copybooks, subprograms, 3270 formats, tickets, runtime logs, and config linked · sources cited

How it works

From Scattered, Undocumented Knowledge to a Queryable Model of Your Application.

Application knowledge lives in silos that do not connect, including source code in one system, incident history in another, and runtime behavior which is not stored anywhere. Squirro + RMJ2 ingests every source into a single knowledge graph, then uses RAG grounded against that graph to answer any question in plain language, cited back to the original artifact, retrieved at query time rather than trained into a model.

01

SOURCES

Every knowledge source connected

Source code, developer chats, ticketing systems, and runtime monitoring ingested simultaneously, regardless of which system or vendor they originated in.

Knowledge Sources Every system the application’s knowledge lives in, ingested at the same time 0 artifacts indexed
SourceSystemStatus
COBOL programs · copybooks · JCL1,284 programs · 3,912 copybooks · 2,106 JCL members GitLab · PDSSynced
Developer chats#mainframe-support · #claims-batch · 41,300 messages TeamsSynced
Tickets & change recordsINC · CHG · 6,730 records since 2014 ServiceNowSynced
Runtime behaviorexecution traces · dynamic SQL · BMP job logs IMS · DB2 monitorStreaming
3270 screen formats & business rules618 BMS maps · 240 rule documents BMS · SharePointSynced
Runbooks & architecture notes1,140 documents · last edited 2011–2026 ConfluenceSynced
Ingested whichever system or vendor each source came from · runtime behavior included, not just static code
02

KNOWLEDGE GRAPH

Knowledge graph structures and links it

Code, tickets, change records, and runtime behavior connected into a single queryable model of how the application actually works, not just what the code says.

Knowledge Graph How the application actually works — code, tickets, change records and runtime behavior in one model PAYCLM01
Linked to PAYCLM01 · online claims payment27 nodes · 6 edge types
PAYCLM01.cbl→ 11 subprograms→ 6 copybooks→ 3 × 3270 formats→ 2 business rules→ BMP job PAYNITE
Edges resolved while indexing6 types
calls · CLMVAL02, CLMPOST, +9 · from the CALL statementsstatic
includes · CLMREC, PAYHDR, +4 copybooksstatic
displays · BMS maps CLM010, CLM020, CLM030static
referenced by · INC-48213, INC-48377 · CHG-2026-0311tickets
runs as · BMP job PAYNITE · 10–15 × daily, 22:00–23:00runtime
writes to · CLAIMS_HIST via dynamic SQL — seen only in execution traces, not in the sourcederived
The dynamic-SQL dependency is in no document and in no one’s head — the graph derived it from runtime behavior
03

GROUNDED ANSWERS

RAG retrieves and grounds every answer

Developers ask in plain English. RAG retrieves the relevant knowledge from the graph and grounds the response in it, cited to the exact source artifact, with no escalation to the one person who knows.

Ask Grounded in the knowledge graph, cited to the source artifact JK · Developer · joined 2026
Which programs write to CLAIMS_HIST, and which batch jobs would a schema change break?

Three programs write to CLAIMS_HIST: PAYCLM01 and CLMPOST through embedded SQL, and CLMARCH through dynamic SQL built at runtime — the last one is visible only in the execution traces.12

A schema change would break two batch jobs: PAYNITE (BMP, nightly 22:00–23:00) and CLMARCHW (weekly archive). Both have documented fallbacks; the archive job has none for a partial run.3

Open ticket INC-48377 touches CLMPOST — coordinate before changing the column list.4
1PAYCLM01.cbl · CLMPOST.cblPROCEDURE DIVISION · EXEC SQL blocks 2Execution trace · CLMARCHdynamic SQL · 2026-09-14 22:41 3PAYNITE.jcl · CLMARCHW.jclrunbook RB-114 · fallback section 4INC-48377ServiceNow · open · CLMPOST
Answered in 12 s · no escalation to the one person who knows · logged to the audit trail
04

CONTINUITY

Knowledge stays current and preserved

Continuous pipelines keep the model up to date as sources change. The knowledge remains permanently, independent of the individuals who held it.

Pipelines Continuous sync — every source change re-indexed, every link re-resolved Live
6Sources syncing
58,412Artifacts in the graph
4 minChange → indexed, median
0Held by one person only
TimeEventSource
22:14Merge !4821 · 3 programs re-indexed, 2 new CALL edgesGitLab
22:15INC-48377 closed · resolution linked to CLMPOSTServiceNow
22:41PAYNITE run · execution trace captured, dynamic SQL re-resolvedRuntime
23:0214 messages parsed · 2 answers linked to map CLM020Teams
FriAccount of R. Haller (retired) archived · 312 authored artifacts remain in the graphDirectory
The knowledge stays when the person goes — nothing in the model depends on who is still on the team
05

EFFICIENCY

Search time reduced, token usage minimised

In most application environments, agents and developers spend more time locating information than producing answers. The knowledge graph collapses that search step – cutting time to answer and sharply reducing the token cost of every query.

Insights Where each query’s time and tokens go, with and without the knowledge graph 0% less context per query
17 sMedian time to a cited answer
6Artifacts retrieved per query, graph-scoped
2.1kContext tokens per query
Locating vs. answering · one typical queryillustrative
Full-text search · 412 chunks read38k tok
Graph-scoped retrieval · 6 artifacts2.1k tok
Escalation to a senior developer2–3 days
Cited answer from the graph17 s
The search step collapses — time to answer and the token cost of every query fall with it

Integrations

Connect Seamlessly with Your Existing Data

Break down data silos and integrate all of your existing data, creating a single source of truth.

Data source connectors Squirro plugs into

Your Expansion Path

Each Stage Deepens the Knowledge Model and Extends Its Reach Across IT Operations.

Each stage increases automation depth and adds intelligence. This is a transformation journey, not a standalone tool.

  1. 01 To start

    Digital assistant for legacy applications

    Knowledge graph across source code, chats, tickets, and runtime behavior. Natural language Q&A with cited answers. Live in 6–8 weeks.

  2. 02 Next

    IT self-service resolution

    The same knowledge model applied to IT support, with common application questions resolved instantly, escalations reduced, senior staff freed for genuinely complex work.

  3. 03 Next

    Application modernization acceleration

    The knowledge graph becomes the foundation for modernization. Architecture teams query dependencies, behavior, and interactions before writing a line of migration code, replacing the stalled pre-study.

  4. 04 Vision

    Full application intelligence platform

    A continuously learning model of your entire application estate, spanning knowledge preservation, support deflection, modernization, and autonomous issue resolution, compounding with every system it ingests.

Our Platform

Connect Once, Govern Once, Reuse Everywhere.

Most enterprises rebuild the same plumbing for every AI project — connectors, governance, retrieval, audit. Squirro builds it once. You extend it.

Squirro Platform Copy-1

One Platform. Limitless Workflows. Ready to Deploy.

Ready to Deploy

Start Where the Friction Is.

Retiring application knowledge, stalled modernization pre-studies, and slow fault diagnosis create technical friction differently in every organization, so we are mapping the most strategic path to value with a handful of technology teams.

Share your details below for early access, a look at how this fits your stack, or a voice in how it gets built. A few quick questions, then, if it makes sense for both sides, a 20-minute conversation. No pitch.

Browse the Catalog