Environment and package management

The notebook runs against the session's bound runtime: the app-managed default (default-python / default-r) unless you bind another one with notebook_bind_runtime — a named environment you created, or one of the user's own detected interpreters. You never activate environments by hand, and you never install packages from inside a cell. Installs happen in the trusted main process through a single tool, manage_packages, and always land in the currently bound runtime. This page is the workflow for getting a package installed and for knowing when a package is not something you can install yourself.

When a package is missing

A run that fails with ImportError / ModuleNotFoundError (Python) or Error in library(x): there is no package called 'x' (R) means the package is not in the environment yet. The fix is one manage_packages call, not a code change. Do not rewrite the cell to use a different library that "does roughly the same thing" — install the package the task actually needs. Do not fall back to reading data or computing results a worse way to dodge the missing import.

Check an installed version

Use inspect_packages(language, packages) when the user asks whether a package is installed in an app-managed runtime or which version is present, or when your code depends on a version-specific feature. It reads package metadata from the session's bound app-managed runtime without importing the package or changing the environment. An installed result does not prove the import will succeed; use notebook_execute when importability itself is the question.

Inspection does not provision a missing app-managed default runtime. If it reports DEFAULT_RUNTIME_NOT_READY, use notebook_execute in that language to prepare the runtime under notebook execution approval, then retry inspect_packages.

inspect_packages intentionally rejects a user-owned external runtime because reading its metadata executes that interpreter. Use notebook_execute for an external runtime so the user sees the normal notebook execution approval.

Do not use inspection as a mandatory preflight for every install. For a clear missing-package error, call manage_packages directly; installation is the recovery action, while inspection is for explicit version and compatibility questions.

Route by language

• Python package → manage_packages(language="python", packages=["numpy", "pandas"]).
• R package → manage_packages(language="r", packages=["ggplot2"]). R packages install from conda-forge as r-<name> automatically; pass the plain CRAN name (ggplot2, not r-ggplot2).
• A PyPI-only Python package that is not on conda → add usePip=true.
• A package that needs a specific conda channel → pass channels=["bioconda"]. Leave channels off otherwise; the app supplies the right default mirror.

manage_packages always returns a target receipt. When resolution succeeds, it identifies the implicit default or explicit binding and the actual runtime; selection:"unresolved" means no target could be established. After a failed bind or switch, inspect bindingChanged and target before deciding where a later install would run. Every install persists — there is no "temporary" install to undo later. Install once; it stays available in later cells and sessions on that runtime. To install into a different environment, bind or switch to it first (notebook_bind_runtime / notebook_switch_runtime); there is no per-call environment argument. The app-managed defaults are additive-only (bare name or name==version); for uninstalls, version ranges, or git/URL specs, create a named environment and install there.

Restart the kernel after an install when told to

manage_packages returns a compact result with ok, needsRestart, the installer method, the target receipt, verified packageChanges, and an actionable error on failure. A requested package change reports installed, updated, unchanged, or removed plus its observed before/after version when available. unchanged means only that distribution metadata in the reported target did not change; it does not prove the current Kernel can import the module. When importability or the loaded version matters, use notebook_execute to import the package and inspect its version. When needsRestart is true (always true for R, because the running kernel holds the old library state), call notebook_restart before you import or library()-load the new package, then re-run the cell. For Python, a fresh import usually sees the new package without a restart; if an earlier failed import was cached, restart and retry.

Never install any other way

These bypass the install gate and are forbidden:

• OS package managers — apt, brew, yum — and sudo.
• curl | bash, downloading and running installers, or hand-rolled subprocess installs.
• In-cell installs: %pip install, !pip install, install.packages(...), remotes::install_github(...). These run inside the kernel, which has no install-network path and is sandboxed in a later phase — they do not belong in a cell.

When to stop and tell the user

Some things are not a manage_packages install:

• A package that needs a system / OS-level dependency (a compiler, a shared C library, a CUDA/GPU toolchain) that is not present. Stop and report the limitation to the user — say what is needed and why it is out of scope here; do not try to self-install system dependencies.

When you choose to select a newly created isolated environment (to remove/downgrade a package, use richer specs, or keep a project's deps separate), await manage_environments(action:"create", language, name), check created.runnable, and use its canonical created.runtimeId. Create does not select the environment; continue only when created.runnable is true. With no existing binding for that language, pass the receipt's runtimeId to notebook_bind_runtime; with an existing binding, pass it to notebook_switch_runtime. Do not use the short environment name as runtimeId, and do not call list merely to rediscover the environment you just created. Only action:"list" returns the full environment snapshot; create and remove do not return that snapshot, and instead return receipts for the environment they created or removed. The app-managed defaults cannot be removed.