- Category
- AI Agents
- Rank
- No. 1972Tools index
- Pricing
- Open Source
- Type
- AGENT
- Builder
- Yeachan-Heo
- GitHub
- 244 stars
- Latest release
- v0.5.1
- Date
About
One-click scientific research lab for OpenCode — seamless .ipynb and REPL integration so your agent can do real data science.
What it does
My-Jogyo organizes agent-led research around goals, hypotheses, experiments, findings, and reports. A planner directs work, a Python executor runs analysis, a critic challenges claims, and a writer turns results into a narrative. Sessions can continue, replay, or branch while notebooks preserve the experimental record.
Why it's ranked here
The appeal is disciplined research automation, not merely code execution. Structured evidence markers, acceptance criteria, checkpoints, persistent sessions, and adversarial verification form a coherent workflow. Multiple entry points also make the system useful from either OpenCode or Claude Code.
What's good
Findings require confidence intervals and effect sizes or reports downgrade them to exploratory observations. Goal checks can evaluate metric thresholds, required markers, artifact creation, and finding counts. Atomic writes, path confinement, symlink rejection, and ordered locks show serious attention to research-state integrity.
Tradeoffs
The workflow imposes substantial ceremony through markers, hypotheses, acceptance criteria, trust checks, and specialized agent roles. Its package declares only macOS and Linux support. Installation also assumes OpenCode or a separately built MCP server, plus Bun for the main package and Python for analysis.
How to use it well
Use it for multi-step data science where experiments, evidence, artifacts, and narrative reports must stay connected. Choose interactive mode while exploring and autonomous mode only when the goal is clear. It does not cover Windows, and it is not presented as a general notebook editor independent of an agent host.
Technical notes+
package.json defines an ES module npm package, a CLI install surface, Bun 1.0 or newer, an OpenCode peer dependency, MIT licensing, and macOS/Linux targets. src/index.ts implements installation with category and path validation, symlink checks, confined config writes, and lock recovery. src/mcp/index.ts creates a stdio MCP server and maps twelve named research tools to handlers. src/lib/goal-gates.ts evaluates notebook acceptance criteria including metrics, markers, artifacts, and finding counts. src/lib/atomic-write.ts uses exclusive temporary files, file synchronization, rename-based replacement, regular-file checks, and no-follow opens. pyproject.toml defines the Python side for Python 3.8 or newer and configures pytest.
Observed
- License
- MIT
- Primary implementation
- TypeScript ES module package with Python bridge support
- Install surface
- npm package, Bun-based CLI, OpenCode plugin configuration, and source-built MCP server
- Interfaces
- OpenCode commands, CLI, and stdio MCP server exposing twelve research tools
- Platform support
- macOS and Linux
- Runtime requirements
- Bun 1.0 or newer and Python 3.8 or newer
Read from README.md, package.json, pyproject.toml, src/index.ts, src/mcp/index.ts, src/lib/goal-gates.ts, src/lib/lock-paths.js, src/lib/lock-paths.ts, src/lib/bridge-meta.js, src/lib/bridge-meta.ts, src/lib/atomic-write.js, src/lib/atomic-write.ts, src/lib/lock-paths.d.ts.
What it can do
Execute Jupyter notebook files
.ipynb files → Executed notebook results
Run interactive Python code
Python code commands → Code execution results
Perform automated data analysis
Raw datasets → Data analysis results and insights
Generate data science code
Research requirements or prompts → Python data science code
Create scientific visualizations
Data and visualization parameters → Charts, graphs, and plots
Process scientific datasets
Raw scientific data files → Cleaned and processed datasets
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.