What File Formats Does DeepSeek-Reasonix Accept as Input?

DeepSeek-Reasonix accepts plain-text, Markdown, JSON, YAML, TOML, CSV, and common image formats like PNG and JPG, supporting both CLI piping and file-path inputs.

DeepSeek-Reasonix is an open-source reasoning engine designed to process diverse data types through its command-line interface and runtime. Understanding what file formats DeepSeek-Reasonix accepts as input is essential for configuring workflows and integrating the tool into your data pipeline. The repository defines specific parsing rules and input handlers across its documentation and internal tool implementations.

Text and Structured Configuration Formats

Plain-Text and Markdown Files

DeepSeek-Reasonix handles plain-text (.txt) as line-oriented input that can be piped directly into the reasonix run command. For rich-text documents, Markdown (.md) preserves headings, code blocks, and lists, with the parser retaining formatting for UI display and prompt generation. These formats are specified in docs/SPEC.md under the Input handling and Markdown support sections.

JSON and YAML Support

The runtime accepts JSON (.json) for structured data, tool outputs, configuration files, and serialized session histories. According to docs/SPEC.md, JSON serves dual purposes as both input data and a source for tool arguments. YAML (.yml / .yaml) provides an alternate configuration format compatible with reasonix.toml overrides and custom skill definitions, documented in docs/GUIDE.md.

TOML Configuration

TOML (.toml) functions as the primary configuration language for the project's reasonix.toml file. This file defines models, pricing structures, and skill folders. The reasonix init command generates this file at the project root, with detailed specifications available in docs/GUIDE.md under TOML configuration.

Specialized Data and Binary Inputs

CSV for Tabular Processing

For tabular datasets, DeepSeek-Reasonix accepts CSV (.csv) files through its dedicated csv tool. As documented in docs/TOOL_APPROVAL_MODES.md, the engine treats each row as a separate record for batch processing and analysis tasks.

Image Files for Vision-Based Skills

When utilizing image-analysis capabilities, the system accepts common raster formats including .png and .jpg. The docs/BOT_GUIDE.md specifies that these binary inputs are handled via base-64 encoding or direct file-path inputs, enabling vision-based skills to process visual data alongside textual content.

Input Methods and CLI Usage

DeepSeek-Reasonix supports multiple input delivery methods: standard input piping, direct file paths, and tool-specific arguments. The implementation in internal/tool/* handles format validation and parsing for each input type.

Examples of valid input methods:

# Plain-text via pipe

echo "Summarize this paragraph..." | reasonix run

# Markdown file

reasonix run --file notes.md

# JSON data for a custom tool

reasonix run --tool myTool --input data.json

# TOML configuration (project root)

reasonix init   # creates reasonix.toml

Documentation and Implementation References

The definitive source for supported formats resides in docs/SPEC.md, which details input schemas and parsing rules. Practical configuration guidance appears in docs/GUIDE.md, while vision-specific inputs are covered in docs/BOT_GUIDE.md. The reasonix.example.toml file demonstrates how to reference various format types in project configuration, and the internal/tool/* directory contains the actual implementations for built-in format handlers.

Summary

• DeepSeek-Reasonix accepts plain-text, Markdown, JSON, YAML, TOML, CSV, and image formats (PNG, JPG) as valid inputs.
• Text and configuration formats are documented in docs/SPEC.md and docs/GUIDE.md, while image handling appears in docs/BOT_GUIDE.md.
• Inputs can be supplied via standard input piping, file paths (using --file or --input flags), or as tool arguments.
• The csv tool processes tabular data row-by-row, and vision skills accept base-64 encoded or file-path image references.

Frequently Asked Questions

Can DeepSeek-Reasonix process multiple file formats in a single command?

While the CLI primarily processes one primary input source at a time, you can chain operations or use configuration files in reasonix.toml to define workflows that reference multiple format types. The internal/tool/* implementations handle format-specific parsing for each distinct input type.

Does DeepSeek-Reasonix support binary file formats other than images?

Currently, the repository explicitly documents support for raster image formats (.png, .jpg) as binary inputs. Other binary formats would require custom tool implementations within the internal/tool/ directory or base-64 encoding through the methods described in docs/BOT_GUIDE.md.

Where is the input format validation defined in the source code?

Input validation and parsing logic are implemented in the internal/tool/* directory, with schema definitions and format specifications documented in docs/SPEC.md. The reasonix.example.toml file provides reference implementations for declaring acceptable format types in project configurations.

Can I use YAML instead of TOML for my main configuration?

While TOML (.toml) is the primary configuration format for reasonix.toml, YAML (.yml / .yaml) is supported for configuration overrides and custom skill definitions as specified in docs/GUIDE.md. You can use YAML for specific tool inputs and skill definitions even when the main project file uses TOML.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}