0.1.0-rp
Release notes for 0.1.0-rp.
This release focuses on getting users from “install” to “run and inspect” with a first working set of CLI commands, clearer onboarding guidance, and an expanding set of MCP-based tracing/visualization capabilities. It adds experiment scaffolding and validation, a matrix-driven experiment runner with dry-run and live watching, and commands to list prior runs and reveal where outputs were written. On the data side, it wires up trace-store access through SQL and MCP paths, and makes scatter-based visualizations consume real trace data. It also expands the platform experience (experiment pages, experiment onboarding, and run viewing affordances), plus updates install/docs to call out key prerequisites (Docker/model-provider API key) and make common onboarding pitfalls clearer.
Docs
- Simplified AXP onboarding pages (Overview/Installation/Getting Started) and smoothed onboarding install/run prerequisites.
- Updated onboarding to explicitly mention Docker is needed before running
axp run. See Installation. - Updated onboarding to explicitly mention the model-provider API key requirement before
axp run. See Environment and Auth. - Clarified
.zshrcbehavior during the AXP onboarding script, including a workaround when shell sessions don’t inherit changes. - Addressed a confusing “no job control” warning shown during non-interactive installs.
- Documented the end-to-end experiment flow: define → run → inspect. See Getting Started, Experiment YAML, and Runs.
- Documented beta-period MCP app integration steps for non-developer users.
CLI
- Added an update-available nudge/banner to the AXP CLI during subcommand usage.
- Added experiment creation and validation commands (
axp create/axp validate) so users can scaffold a runnableexperiment.yamland quickly lint/edit it. See AXP CLI. - Added
axp runmatrix support, including--dry-run, parallelism via-j, and experiment variant handling via--variant. See Runs. - Added support for repeating experiment runs for statistical analysis via a
--trials/repeat-style workflow onaxp run. - Added
axp run --watchto tail live events while a run is executing. - Added
axp listto enumerate recent runs. - Added
axp resultsto print output paths for a run. - Added
axp run --dry-runhandling improvements so scratch directory management is RAII-managed (avoiding the previous manual forget pattern). - Added an “Export to Claude” action on run details that copies a templated prompt to the clipboard for interactive exploration.
Platform / App
- Shipped an experimental experiments page in the platform app.
- Added a basic “get started” home page that shows how to install and begin using AXP.
- Added an experimental “experiment page” experience in the platform app.
- Added an end-to-end onboarding path for MCP-related integration during the beta period (as documented above).
- Added an end-to-end “login + upload + runs dashboard” scaffold to drive the runs experience end-to-end.
- Shipped an “update” to platform marketing pages, including a legacy link/button.
MCP / Tracing / Analytics
- Added MCP infrastructure with a remote endpoint and authentication.
- Added an MCP tool path that connects to ClickHouse so scatter visuals can consume real trace data.
- Added a read-only SQL-style query interface over the trace store through the MCP tool path.
- Ported scatter plot widgets into the MCP app (including the cost-score scatter plot).
- Added auth injection support for experiment-declared secrets.
Experiment / Example Content
- Added
axp runexperiment workflows to support iteration-focused variants and repeated runs to validate bias vs stochastic variance (example-focused experiment variants and higher-scale/locked-constraint variants). - Added a
descriptionfield to experiment definitions.
Developer Experience / Infra
- Set up Linear release tracking and release notes generation.
- Added a webhook-style connection from Linear to GitHub when a new release is cut.
- Created the AXP CLI distribution release workflow.
- Added an MVP way to upload run results to hosted object storage.
- Updated CI toolchain versions (Rust/tooling in CI).
- Added local ClickHouse support for AXP local development.
- Started foundational core app scaffolding and rust dev infra (tests/lint/CI floor).
- Added an experiment run “experimental” flag.