Claude Code sessions are now first-class in your project: encrypted transcripts sync across machines, resume exactly where you left off on any of them, and every past session is searchable — from the web and the terminal. This is the Sessions preview shipping for real, phases 1 through 3.
Highlights
- First-class, encrypted sessions — sessions live in a Sessions tab and link to tasks. Full transcripts are encrypted at rest with AES-256-GCM under the same key infrastructure as the Secret vault; they are never stored in the clear.
- Cross-machine sync + exact resume — the CLI session watcher streams your active transcript up as you work.
arcanova sessions resume <SES-n> pulls it down on another machine and relaunches claude --resume so you continue the same conversation, not a summary of it.
- Resume forks, never collides — resuming a synced session on a second machine freezes the original and continues in a new linked session, so two machines can never scribble over one transcript. The lineage is recorded and shown, so a resume chain reads as one thread.
- Search past sessions, web + CLI — a search bar in the Sessions panel and the new
arcanova recall "<query>" answer "what do past sessions know about X?". arcanova recall --files <path> lists every synced session that touched a file or directory, with per-session action summaries.
- Recall for agents — the
/arcanova:recall skill lets a Claude Code agent consult prior sessions before it plans, and the autopilot RESEARCH step now weaves prior decisions, failed approaches, and relevant files into its research brief. Recall failing or an index being off never blocks a task — it degrades to a one-line notice.
- Touched-files index — "every session that modified file X" is surfaced where you already look: on the session, the task, and the Files tab.
- Handoffs + auto-checkpoints —
/arcanova:handoff stores a distilled goal / decisions / current-state / next-steps doc on a session, and the watcher writes rolling auto-checkpoints, so cross-device pickup works even when an exact transcript resume isn't available.
What is searchable (and what stays private)
Transcripts stay encrypted. To make sessions searchable we index only session titles, repo-relative file paths, and short secret-scrubbed previews of prompts and responses (each preview is run through the CLI's secret scanner and capped before it ever leaves your machine). Tool inputs, tool output, command output, file contents, and diffs are never indexed — they exist only inside the encrypted transcript. Privacy-sensitive projects can turn the index off per project and still sync; search then falls back to titles and machine names.
Upgrade notes
The sessions transcript upload gained new fields, so the protocol version bumped.
Upgrade: rerun npx arcanova-cli@latest (this ships 0.5.3) to pick up arcanova sessions, arcanova recall, and arcanova session-watch. Whenever a changelog entry says the protocol bumped, rerunning that one command is the fix — npx caches an old CLI until you do.