DSAgt

DataSmith Agent, an AI-assisted data pipeline builder.

DSAgt connects an MCP-compatible AI coding agent to code registration, a semantic knowledge base, skills discovery and creation, execution provenance, and observability infrastructure. It exposes these capabilities to a user's existing agent CLI or VS Code extension (Claude Code, OpenCode, Codex, and others).

Prerequisites: Python 3.12 or later (CI tests 3.12 and 3.13) on Apple Silicon or Linux x86_64 (onnxruntime, which runs the local embedding model, has no Intel Mac wheel), and one of the agent platforms below, installed and authenticated against the LLM provider you intend to use.

Agent	Install	Verify
Claude Code	npm i -g @anthropic-ai/claude-code	claude --version
Goose	See Goose docs	goose --version
Codex	npm i -g @openai/codex (or brew install --cask codex)	codex --version
opencode	See opencode docs	opencode --version
Cline	npm i -g cline	cline --version

Installation

python3.13 -m venv ~/.venvs/dsagt          # or: conda create -n dsagt python=3.13 && conda activate dsagt
source ~/.venvs/dsagt/bin/activate         # (Windows venv: ~\.venvs\dsagt\Scripts\activate)
pip install "git+https://github.com/AI-ModCon/dsagt.git"
dsagt --version                            # 0.2.1

This puts the dsagt CLI on your PATH. Create your first project. dsagt init is interactive (it prompts for the agent platform, project location, packaged knowledge collections, and skill sources) and sets up the knowledge base on first run:

dsagt init                      # interactive; pick agent, collections, sources

Then start dsagt (shorthand for starting your agent with the dsagt MCP server enabled), or open the project in VS Code:

dsagt start <my-project>   # ≈ cd ~/dsagt-projects/my-project && claude   (or your preferred agent)

With a VS Code agent extension, open the folder as a project in VS Code and start the agent; dsagt init has already written the dsagt MCP server into the agent's native config (for Claude, the project's .mcp.json).

To upgrade later, reinstall; re-running dsagt init reconfigures an existing project in place:

pip install --upgrade "git+https://github.com/AI-ModCon/dsagt.git"

Pin to a specific release: e.g. pip install "git+https://github.com/AI-ModCon/dsagt.git@0.2.0".

For a development install from a clone, see the Developer Guide.

Capabilities

Capability	Description
Code Registry	Register CLI codes as markdown specs; the agent discovers them via search_registry and runs them from its shell
Knowledge Base	Hybrid semantic + keyword (BM25) search over indexed ChromaDB collections
Skills Discovery	Search the external skill corpus and install workflow skills on demand via search_skills / install_skill; an uninstalled skill takes no space in the agent's context
Provenance	dsagt-run wrapper records every code execution to trace_archive/; reconstruct_pipeline renders it as a runnable script
Explicit Memory	User-confirmed facts persisted to YAML and the knowledge base
Episodic Memory	Opt-in: the MCP server chunks and embeds each session turn into a searchable session_memory collection (recency-weighted retrieval)
Observability	Serverless MLflow tracing (a per-project SQLite file): DSAgt's own spans plus agent traces recovered from the on-disk transcript

The Quick Start exercises all of these in a single session.