Hub guide

rawctx workflow docs

rawctx packages your company meaning into checksummed, versioned packages for AI agents, analytics, and code.

Quickstart

3-minute quickstart

Install rawctx, review a package, download a published snapshot, and load the same package context from Python or your agent workflow.

PyPIrawctx CLI

Default surfaceHub

Download pathsnapshot-download

Registry hosthttps://api.rawctx.dev

Open search Request a pilot Open Hub home

python3 -m pip install --upgrade rawctx
rawctx search "stripe subscriptions" --sort recent
rawctx snapshot-download @pasar6987/stripe-subscriptions
rawctx to-prompt @pasar6987/stripe-subscriptions --datasets subscriptions,invoices --max-tokens 1200

Start with the problem

Move from review to reusable package context

These docs mirror the product story: review provenance first, load a trusted package second, and only publish from the scope your team actually owns.

Review

Review packages by business meaning, source system, or domain workflow.

Use web or CLI to see what a package is, where it came from, and whether it is published or still a candidate.

Open Review guide

Claim

Use the right ownership path for published and candidate packages.

Published packages load directly in rawctx. Candidate packages stay review-first and point back upstream until you publish a version you own.

Open Claim guide

Publish

Publish the version your team actually owns.

Keep provenance, versioning, and release boundaries explicit before the package becomes the shared source for code and AI.

Open Publish guide

Compare

Diff artifacts before you rely on them in code or AI flows.

Use semantic and prompt diffs for release gates, migrations, and package review handoffs.

Open Compare guide

CLI-first

Keep agent handoff shell-native.

Use stdout, JSON, and exit codes instead of binding rawctx to one host-specific tool protocol.

Open CLI-first guide

Auth

Know when review stays public and when workspace auth matters.

Search, inspect, and many downloads stay public; claim, publish, settings, and private access do not.

Open Auth guide

Workspace

Operate invites, tokens, and ownership in the right workspace.

Use rawctx Hub for review and the workspace for the settings and release boundary you own.

Open Workspace guide

Formats

Normalize deliberately between OSI and native MetricFlow.

Convert when you need canonical OSI review, or publish native when the dbt snapshot is the product.

Open Formats guide

Package availability

Published package means rawctx can hand off directly to snapshot-download, load(), and to_prompt() for published OSI or native MetricFlow snapshots. Candidate packages stay reviewable in rawctx Hub, but ownership and runnable access stay upstream until a maintainer publishes a version they own.

Authentication boundary

Public search, info, review, and many downloads do not need login. Login is reserved for publish, claim, favorites, self-serve workspace creation, settings, or private workspace access.

Automation and Python

The CLI and Python SDK share the same public-first model, so notebooks and scripts can use the same review flow plus typed model loading, native MetricFlow snapshots, and prompt compression you see in the docs.