Skip to content

Upgrading to TokenPak 1.25.1

python -m pip install --upgrade "tokenpak==1.25.1"
tokenpak --version
tokenpak doctor

Keep the previous environment and back up configuration and persistent state before changing a running installation. Existing custom configuration and pricing rows are preserved; a version upgrade does not authorize rewriting stored costs.

Changes to check

  • tokenpak savings --verify compares the byte estimator with an independent tokenizer on a packaged corpus. Install tokenpak[tokens]==1.25.1; offline use also needs its encoding data cached. Stored prompts are not reconstructed.
  • Where vault injection is enabled, retrieval has a configurable 2,000 ms deadline and two worker slots. Timeout or backlog forwards the original bytes and records a degradation reason. Configure TOKENPAK_RETRIEVAL_TIMEOUT_MS or vault.retrieval_timeout_ms as needed.
  • Role-bearing conversation turns remain intact across companion and Pak-builder processing. Concurrent native launches use separate generated files; custom hooks are composed and customized skills are preserved during managed upgrades.
  • Session displays distinguish observed usage, estimated cost and burn, guard runway, and session remainder. Mixed model/effort history and unsupported effort values remain unavailable for homogeneous forecast calibration.
  • Corrected seed catalogs carry explicit unit and provenance metadata. Existing legacy rows keep their numeric interpretation until an explicit refresh, and stored costs are not rewritten. The reprocess pricing-version override remains ineffective; do not use it as evidence of historical repricing.

Optional Pro daemon

Pro 0.2.0 declares support for exactly OSS 1.25.1 and TIP-1.0. Upgrade that pair together through the licensed package index. The installed package metadata and daemon handshake must agree before readiness is reported as active. Missing, malformed, unconfigured, or incompatible declarations produce a diagnostic. Compatibility does not grant a license or bypass entitlement checks.

The Pro observation and evaluation foundations retain unavailable-input states. They do not establish measured production reroute policies, admitted empirical priors, human comprehension results, or default-on automation.

Rollback

python -m pip install --upgrade "tokenpak==1.24.0"

For a Pro installation, restore the prior Pro 0.1.6 / OSS 1.24.0 pair together from the retained environment and configuration backup. Preserve private observations and adverse records. Recheck version, doctor, proxy health, and the integrations used by your installation after either upgrade or rollback.

Version 1.25.0 was tagged but never published after a release test dependency failure. Upgrade to the corrected 1.25.1 package; the prior published rollback version remains 1.24.0.