T-FLEX DOCs 15 API

LLM-friendly knowledge/API layer generated from the local T-FLEX DOCs 15.4.0.28 documentation set.

The repository is designed for source-code search, RAG, AI agents and the companion T-FLEX-DOCS-15-MCP read-only MCP server.

Tags

t-flex-docs tflex pdm plm dotnet-api chm jsonl markdown llm rag mcp

Contents

• raw/
  • DOCsAPI.chm — .NET API reference supplied with T-FLEX DOCs.
  • DOCs.chm — end-user and administration help.
• llm/symbols.jsonl — compact API-symbol reference, one symbol per JSON line.
• llm/page_index.jsonl — compact searchable index of every API/manual page, with pointers to full-page shards.
• llm/pages/*.jsonl — normalized full documentation pages split into Git-friendly shards.
• llm/types/*.md — complete Markdown pages grouped by declaring .NET type.
• llm/page_locations.jsonl — page ID to shard/line mapping.
• llm/manifest.json and llm/pages_manifest.json — counts, source metadata and shard checksums.
• graph/static_nodes.jsonl — documentation/API graph nodes.
• graph/edges/*.jsonl — internal links and type-to-member containment edges.
• graph/capabilities.seed.jsonl — curated DOCs capability map for agent routing.
• scripts/convert_tflex_docs.py — reproducible CHM-to-dataset converter.
• scripts/cab_concat.c — small libarchive helper used for CHM LZX decompression on Linux.

Generated dataset

Dataset	Count
Documentation pages	37,043
API symbols	29,538
Declaring types / Markdown pages	3,763
Graph nodes	37,043
Graph edges	463,658
Capability seeds	15

Symbol kinds:

Kind	Count
Methods	11,525
Properties	9,256
Types	3,763
Fields	3,170
Overload groups	1,398
Namespaces	300
Events	125
Project root	1

Assemblies represented in the source help include TFlex.DOCs.Model, TFlex.DOCs.Model.Processes, TFlex.Model.Technology, TFlex.DOCs.ProjectManagement, TFlex.DOCs.Common and TFlex.Reporting.

Record formats

A symbol record contains fields such as:

{
  "id": "M:TFlex.DOCs.Model.ServerConnection.Connect(System.String)",
  "kind": "method",
  "assembly": "TFlex.DOCs.Model",
  "namespace": "TFlex.DOCs.Model",
  "type": "TFlex.DOCs.Model.ServerConnection",
  "name": "Connect",
  "qualified_name": "TFlex.DOCs.Model.ServerConnection.Connect",
  "signature": "...",
  "summary": "...",
  "page_shard": "llm/pages/api-004.jsonl",
  "page_line": 1234
}

A compact page-index record contains page metadata, a search excerpt and the exact shard/line pointer required to retrieve the full page.

Recommended LLM usage

• Use llm/symbols.jsonl for API member lookup and code generation.
• Use llm/page_index.jsonl for fast help/manual search.
• Resolve the returned page_shard and page_line only when full context is required.
• Use llm/types/*.md when the model needs an entire class, interface, enum or delegate context.
• Use graph/edges/*.jsonl for dependency traversal and declaring-type containment.
• Use graph/capabilities.seed.jsonl to map user intentions such as references, files, access rights, workflows and reports to relevant API symbols.

Regeneration

Linux prerequisites:

sudo apt-get install python3 build-essential libarchive-dev

Put the original CHM files in raw/, then run:

python3 scripts/convert_tflex_docs.py \
  --raw raw \
  --out . \
  --work .work \
  --shard-size-mb 24

The converter:

1. parses the CHM directory directly;
2. reconstructs the reset-aligned LZX stream as a CAB container;
3. uses libarchive to recover the uncompressed content stream;
4. normalizes HTML and Windows-1251 text;
5. builds symbol, page, Markdown, graph and manifest artifacts;
6. corrects inconsistent CHM container metadata using a complete two-pass type index.

Intermediate CAB and decompressed stream files remain in .work/ and are excluded from Git.

Validate generated counts, checksums, symbol/type hierarchy, page pointers and graph endpoints with:

python3 scripts/validate_dataset.py .

The same validation runs in .github/workflows/validate.yml.

Repository size

The two original CHM files and generated indexes are intentionally committed as regular Git files. Every individual file is below GitHub's 100 MiB object limit; generated JSONL data is split into bounded shards.

Rights and scope

The conversion code is licensed separately in LICENSE-CODE. That license does not relicense the T-FLEX DOCs documentation, CHM files, product names or generated documentation content. See NOTICE.md.