developer tools

---

id: developer-tools title: Developer Tools tags: [python, cli, mcp, ocr, llm, openai, benchmark] created: 2026-04-11 updated: 2026-04-22 freshness: fresh

Developer Tools

Summary

The Developer Tools category spans 8 repositories covering code intelligence (GitNexus Fleet), AI-powered documentation and diagram generation (AutoGen Doc Generator, AutoGen Mermaid Generator), document format conversion (OllamaPDF2Markdown, Word-to-Markdown Converter, Nemotron Parse on DGX Spark), repository packaging (RepoBundle), and media utilities (Animated GIF Creator). The AutoGen-based tools share a common pyautogen 0.3.1 foundation; the format converters address the recurring need to move content between formats for LLM ingestion pipelines and now span Apple-Silicon (Ollama) and NVIDIA-Blackwell (Nemotron Parse) backends.

Key claims

• GitNexus Fleet indexes repositories into KuzuDB graph databases capturing functions, classes, files, call graphs, imports, and community structure. ^src:gitnexus-fleet
• The tool runs in three phases: parallel cloning (5 workers), parallel indexing (3 workers via GitNexus), and MCP server + HTTP dashboard startup. ^src:gitnexus-fleet
• Queries are served via MCP (stdio protocol) supporting query, impact analysis, context retrieval, and Cypher graph queries. ^src:gitnexus-fleet
• AutoGen Doc Generator uses pyautogen 0.3.1 GroupChat with specialized agents (code analyst, documentation writer, quality reviewer) to auto-generate comprehensive documentation from any codebase. ^src:autogen-doc-generator
• AutoGen Mermaid Generator uses a similar GroupChat pattern with a coder agent, manager agent, and Autogen manager to generate Mermaid diagrams from source code with quality-gated iterations. ^src:autogen-mermaid-generator
• OllamaPDF2Markdown converts PDF pages to images then processes them through Ollama multimodal models (Mistral Small 3.1 24B) with intermediate saves every 5 pages for crash resilience. ^src:ollama-pdf2markdown
• RepoBundle serializes an entire Git repository (text + binary files) into a single human-readable text file using only Python stdlib — no external dependencies. ^src:repobundle
• Word-to-Markdown Converter preserves heading hierarchy, bullet lists, and table structures when converting .docx to Markdown. ^src:word-to-markdown-converter
• Animated GIF Creator supports both image-folder-to-GIF and MOV-to-GIF workflows with automatic resizing and configurable frame duration. ^src:animated-gif-creator
• Nemotron Parse on DGX Spark wraps NVIDIA's <1 B-parameter document-understanding model (ViT-H C-RADIO encoder → mBART decoder) on Grace Blackwell GB10, exposing both transformers and vLLM backends through a single postprocessing path. ^src:nemotron-parse-spark
• The Nemotron v1.2 prompt requires exactly four directive tokens after </s><s>; the harness pins them in DEFAULT_PROMPT and guards the constants with a TestPromptConstants test, since dropping any one silently degrades output structure. ^src:nemotron-parse-spark
• The harness's image-sizing contract enforces the model card's 1024×1280 min / 1664×2048 max envelope through a single iter_document_images entrypoint — bypassing it produces silent quality degradation. ^src:nemotron-parse-spark

Open questions

• Could GitNexus Fleet's graph indices feed into the org wiki for automated cross-repo dependency mapping?
• Could AutoGen Doc Generator + AutoGen Mermaid Generator be composed into a single documentation pipeline?
• Is OllamaPDF2Markdown superseded by the Qwen-based OCR tools in local-inference-mlx for Apple Silicon users — and by Nemotron Parse on the NVIDIA side?
• How do Nemotron Parse's bounding-box + class outputs compare with DeepSeek-VL2 and Qwen3-VL on the same arXiv / NIST sample set?

Related

• local-inference-mlx (complementary local AI tooling; OllamaPDF2Markdown overlaps with Qwen VL OCR)
• agentic-frameworks (AutoGen tools share agent orchestration patterns; potential integration with DeepAgent Azure CLI)

Articles

• timeline — 2025-03-25. ^src:zero-trust-ai-coding-boom

Sources