Concho Launches a Codebase Knowledge Layer for AI Agents
Concho AI launched a system that scans complex enterprise applications and builds a queryable model of architecture, dependencies, workflows, and embedded business rules. The company says the resulting context can be exposed to external AI assistants through the Model Context Protocol, while SiliconANGLE reports that customers are using it to investigate legacy systems and support modernization work. The launch is real, but the vendor's productivity and completeness claims have not been independently benchmarked. LDS proposes a practical evaluation: compare retrieved rules with expert-reviewed ground truth, measure dependency coverage and citation precision, test stale-index behavior, and require a human-approved change plan before an agent modifies mission-critical code.
What happened
Concho AI launched a system designed to turn complex enterprise applications into a queryable knowledge layer for developers, business teams, and AI agents. The company says the product scans code and related technical evidence, models architecture and dependencies, and extracts workflows and business rules that may be scattered across repositories and documentation.
The product exposes this context through the Model Context Protocol so external assistants can query it. SiliconANGLE independently reported the launch and interviewed the company's technology chief about its use in legacy-system modernization. The reporting confirms the product and its intended workflow, but it does not independently establish the vendor's claims about completeness, productivity, or migration outcomes.
Technical context
Repository retrieval is not the same as system understanding. A useful application model must preserve links between code, configuration, data schemas, runtime integrations, ownership, and business intent. It also needs a way to show which evidence supports each generated diagram, rule, or migration recommendation.
| Evaluation layer | Useful measure | Risk to test |
|---|---|---|
| Code coverage | Repositories, languages, and generated artifacts indexed | Important logic omitted from the model |
| Rule extraction | Precision and recall against expert-reviewed rules | Plausible but invented business meaning |
| Dependency graph | Verified links across services and data stores | Missing runtime or configuration edges |
| Freshness | Time from repository change to model update | Agents using stale architecture context |
| Agent use | Task success with traceable evidence | Confident edits outside approved scope |
For practitioners
A pilot should begin with a bounded application whose architecture and critical rules are already known. Reviewers can create a ground-truth set, hide part of it from the system, and then measure retrieval, rule extraction, dependency mapping, and citation quality. Teams should also test deleted code, generated code, feature flags, and cross-repository calls.
Security controls matter because an application model may concentrate sensitive design and business knowledge. Access should follow repository permissions, retrieved evidence should be auditable, and external assistants should receive only the minimum context needed for a task.
Editorial analysis
LDS sees the launch as part of a broader shift from chat-based code search toward persistent, structured context for software agents. The useful question is not whether a graph can be created, but whether teams can prove that it is complete enough, current enough, and precise enough for a specific modernization decision.
What to watch
Watch for reproducible customer benchmarks, supported languages and repository systems, permission boundaries, update latency, evidence-level citations, exportability of the application model, and measured changes in defect rates or modernization lead time.
Key Points
- 1Concho launched a system that models enterprise application architecture, dependencies, workflows, and business rules for human and agent queries.
- 2The product can expose its application context to external AI assistants through the Model Context Protocol, according to the company.
- 3LDS recommends ground-truth coverage, citation precision, freshness, permissions, and change-safety tests before relying on it for modernization.
Scoring Rationale
An impact score of 6.0 reflects a relevant enterprise-agent launch with independent confirmation, tempered by vendor-led evidence and missing reproducible benchmarks.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
