Back to devlog

Blackcrab 0.1.3: steadier sessions and cleaner transcripts

A focused quality release for the workflows people touch every day: switching between conversations, reading long transcripts, and keeping local Claude Code sessions connected reliably.

A focused quality release

Blackcrab 0.1.3 is not a broad feature drop. It is the kind of update that makes the existing app feel calmer under real use.

The release tightens conversation switching, makes transcripts easier to scan, cleans up Claude session startup, and adds anonymous update health events so future releases can be measured more clearly.

Conversations keep their place

The biggest fix is conversation continuity across mode switches. When you move from grid mode into a single conversation, then back to the grid, the app now keeps better track of which backend panel owns each conversation and which transcript should receive new events.

That matters most when a response is still running. Earlier builds could leave the grid showing an older snapshot, or cause the next single-mode open to hit a warning that the session was already being used somewhere else. Blackcrab now tracks panel ownership and transcript state more deliberately so active sessions can continue without fighting over the same saved conversation file.

Transcripts are easier to scan

Long Claude Code sessions can get noisy fast. Tool calls, command output, thinking blocks, file reads, patches, and permission flow all compete with the actual conversation.

In 0.1.3, transcript blocks use more compact drawers. Tool output folds by default when it is not an error, while failures stay visible. Summaries show useful context such as paths, command descriptions, line counts, task counts, and whether a background command is involved.

The goal is simple: you should be able to skim a session, find the important parts, and expand the raw details only when you need them.

Claude session startup is cleaner

This release also tightens how Blackcrab prepares Claude Code sessions. The app refreshes Claude auth state before spawning a session, while avoiding direct injection of keychain access tokens into spawned processes.

That keeps session startup aligned with the local Claude Code CLI and reduces the amount of credential handling Blackcrab needs to do itself.

Update health is now measurable

Blackcrab can now send anonymous app and updater events to the landing site's analytics endpoint. These events cover app launches, update checks, update starts, completed updates, and update failures.

The setting is visible in Settings and can be turned off. The privacy docs and README were updated to explain what is sent and why: the data is meant to help understand whether releases are installing successfully, not to inspect local sessions or Claude traffic.

Upgrade notes

There are no manual migration steps for 0.1.3. Users on 0.1.2 should be able to install the update normally through the app or download a fresh installer from GitHub Releases.

As usual for an early desktop build, the release is still macOS-first, with GitHub Actions also producing Windows and Linux artifacts for testing.

More devlog

Release notes

What changed in Blackcrab 0.1.1 and 0.1.2

The first two updates after the initial preview focused on making Blackcrab more reliable as a daily Claude Code workspace: better attention signals, safer handoffs, usage dashboards, project reporting, faster search, and signed update artifacts.

Design note

Why Blackcrab is a desktop app

The product is not trying to hide Claude Code. It gives the CLI a native workspace for parallel sessions, local context, and the parts of agent work that are easier to manage visually.

Product devlog

Designing the tiled session grid

Running one agent at a time is simple, but it leaves a lot of attention unused. The grid is Blackcrab's attempt to make parallel agent work legible without turning it into a control room.

Engineering

What local-first means for Blackcrab

Local-first is not a vibe for this product. It is a practical constraint around where transcripts live, how Claude Code runs, and what kind of trust the app should ask for.