CancellationPolicy.csC#
42
43
if (order.Status == Shipped)43
return Reject();Roslyn · condition analysisL42–43
FROM CODE TO BUSINESS CONTEXT
Your code.
Explained.
Go beyond what the code says. Understand why it works that way. Connect your repositories. Explore business rules, relationships, and the code behind every answer.
Roslyn + Tree-sitterA MEUSOFT product
ANALYSIS / 001CODE → CONTEXT
condition + code context
R.014BUSINESS RULE INFERENCE
A shipped order
cannot be cancelled.
CancellationPolicy.cs 42–43 ↗
✓An answer.
With its source.↳
With its source.↳
ILLUSTRATIVE PRODUCT EXAMPLEC# / ROSLYN EXAMPLE
Connected code.
Visible understanding.
Git + LocalYour sources
Roslyn / Tree-sitterProject-level analysis
QdrantSemantic search
PostgreSQLData & relationships
01 / THE WORKSPACE
One question.
A traceable answer.
Finding information in code is one thing.
Seeing what supports the answer is another.
Pick an example. Follow the evidence.
INTERACTIVE PRODUCT EXAMPLEIllustrative interface and data · Not the live application
?
When can an order not be cancelled?
Preset scenario · No real LLM calls
Conditions become readable rules.
Each inference includes its source file and line range.
A business rule is not a fact independent of its source.
LLM-derived text is stored and evaluated with its code reference. Select a rule to inspect the sample source code.
Code condition→LLM inference→Rule + code reference→Qdrant
See the process. Not just the result.
From source ingestion to vector storage.
catalog-service05 / 06 STAGE
ANALYZERTree-sitterSandbox
Chat becomes available after the project’s required indexing steps succeed. These statuses are fixed example data.
A workflow may span repositories.
Manage multiple projects and the sources connected to each one.
ANALYZER
RoslynC# / .NET
Project configurationSwitch projects to see that project’s example analyzer, sources, and provider profiles.
Commerce PlatformReady for chat
order-serviceGit repositorymain
payment-serviceLocal project · ~/projects/payment-servicelocal
The two providers are configured independently.
These controls only change the example. Nothing is saved to a server or sent to a provider.
EXAMPLE QUESTIONS
Select a source reference.“How do we know?”
That has an answer, too.
Read an answer alongside the relevant files and lines of code. Go beyond the summary. Check the source.
From an if block
to a business rule.
Evaluate structural information from the project’s Roslyn or Tree-sitter analyzer alongside LLM-derived business explanations.
Evidence is one thing.
Assumptions are another.
Separate source-backed behavior from the model’s interpretation. Missing context should never look like a certain answer.
02 / FROM SOURCE TO ANSWER
Before the answer,
build the context.
Embeddings are only part of the story.
Structure, rules, and relationships
make the code worth talking to.
- 01
Connect your sources.
Bring Git repositories and local projects into the same project context.
- 02
Extract code and business rules.
The project configuration selects Roslyn or Tree-sitter. Selected conditions and their code context feed LLM business rule extraction.
- 03
Index. Track the progress.
Embeddings and rule records go to Qdrant. Projects, jobs, and relationships live in PostgreSQL; progress is visible in Jobs.
- 04
Ask. Trace it back.
Once the required steps succeed, talk to your code. Inspect references, assumptions, and an optional flow diagram together.
CONTEXT MAPEXAMPLE / .NET · ROSLYN
03 / ARCHITECTURE
Each part.
A distinct purpose.
Code structure, semantic similarity, and
relational context are different tools.
ChatToCode brings them together.
[01]
Roslyn
C# / .NETStructural analysis of symbols, methods, and conditions in C# / .NET projects. The analyzer is selected in the project configuration.
INPUT
source.cs→OUTPUTsymbols + conditions[02]
Tree-sitter
PROJECT-BASED PARSINGSyntax parsing suited to the project’s language and configuration. Code structure and conditions provide source context for business rule extraction.
INPUT
project sources→OUTPUTsyntax + conditions[03]
Qdrant
VECTOR STORECode fragments and business rule explanations with embedding-based retrieval. Always linked to the relevant code references.
INPUT
code + rules→OUTPUTsemantic matches[04]
PostgreSQL
DATA & RELATIONSA shared data foundation for projects, repositories, jobs, and graph-style relationships. Keep the pieces of context connected.
INPUT
entities + links→OUTPUTconnected contextYOUR CHOICE OF MODELSChange the model.
Change the model.
Not the approach.
Configure the chat LLM and embedding provider separately. Choose the model that fits the project and the task.
Changing an embedding model may require an index compatibility review and reindexing.
CONFIGURATIONINDEPENDENT PROVIDERS
Chat LLMQuestion → explanation⇄
EmbeddingContent → vector⇄
ChatToCodePROJECT CONTEXT
Same project. Different provider choices.
04 / COMMON QUESTIONSA few important
A few important
details.
The boundaries are
part of the product.
Is ChatToCode a coding agent?
Its focus is understanding existing code and business behavior, not generating code. It combines source analysis, business rule extraction, referenced answers, and flow visualization.
Can it work across multiple repositories?
The product supports multiple projects, each with its own repositories. Git and local sources are managed in the relevant project context. Cross-repository relationships can only be identified where the available code and context support them.
How is the analyzer selected?
The analyzer is selected per project: Roslyn or Tree-sitter. Projects do not all have to use the same engine. Available languages and analysis depth depend on the project configuration and the application’s support.
Are extracted business rules established facts?
No. An LLM-generated explanation is an inference. Evaluate it alongside its code reference. ChatToCode separates source-backed information, assumptions, and missing context. A source reference alone does not make an interpretation error-free.
Which model receives my code?
That depends on the selected embedding and LLM providers and the application configuration. With a remote provider, processed code or context may be sent to that service. Review the provider’s data processing terms during setup.
Can I connect my repository on this page?
This is a static product site. Interactive screens use illustrative data; they do not access Git, build indexes, or call real LLMs. Get in touch to discuss the application and your use case.
THE NEXT QUESTION IS YOURSWhat’s inside
What’s inside
your code?
Let’s talk about ChatToCode in the context of your own work.