Skip to content

Changelog

What an agent sees: tools, their arguments and results, prompts, playbook topics, skills and the skill packs. Tool names have no aliases: a name that is not in the tool catalog gets "unknown tool".

0.8.0 — ships with ParleML 0.8.0

The first release of ParleML's agent layer.

Tools. 26 MCP tools at /mcp, each held to one scope (Tool catalog):

  • 17 read tools (analyst): the playbook, workspaces and their objects, the workspace brief, the catalog, SQL over the lakehouse, table profiles, EDA, AutoDoc, capabilities, model evaluation, insights, rules, chart data, run progress, the platform's settings and templates, and comparisons.
  • 3 compute tools (compute): probe_dataset, what_if and simulate_decision.
  • 6 builder tools (builder, bound to workspaces and a run budget): prepare_data, author_recipe, run_experiment, score_batch, register_model and update_workspace.
  • No tool promotes a model, deletes anything or changes settings (Security model).

Resources and prompts. Four parleml:// resources and seven prompts, such as audit_model and deliver_model (Resources & prompts).

Apps. 17 MCP Apps with 51 interactive panels that render tool results inline in hosts that support them (Apps).

Playbook. 20 topics the agent reads through get_playbook: the iron rules, the decision ladder, the API, explainability, charts, errors, delivery and iteration, a schema per experiment family and a playbook per experiment type (Overview).

Skills. 17 skills: seven parleml-* analyst skills and the ten ml-* lifecycle skills, packaged for Claude Code, Cursor, Codex and Gemini CLI from one source (Skills & packaging).

Agent sessions. ParleML runs Codex CLI, Gemini CLI, Claude Code or Cursor on its own server to deliver, investigate or operate a model in one workspace, and asks people every decision that is theirs (Agent sessions).

Connect. Claude Code, Cursor, VS Code, Codex and Gemini CLI over HTTP with a token; Claude Desktop through a local stdio server; claude.ai and ChatGPT through OAuth (Connect your agent, Tokens & scopes).

Next