sctxx: pricing, features, alternatives

sctxx (Session ConTeXt eXtractor) is a local CLI that turns a finished coding-agent session into a warm start for another agent. The repo handyutils/sctxx is Apache-2.0, created 2026-09-10, language Rust. On 2026-09-13 GitHub listed 18 stars and 13 forks. Latest release v0.3.0 on 2026-09-11. crates.io listed version 0.3.0 and 55 downloads the same day. Docs: handyutils.github.io/sctxx. It is not a hosted memory product.

Compare ccusage if you wanted cost reports from the same logs, Claude Code if you needed the host, or OpenCode if you wanted another agent CLI.

Key Features

Three stores: Claude Code ~/.claude/projects/, Codex CLI ~/.codex/sessions/ (including .jsonl.zst), Pi ~/.pi/agent/sessions/ v1 to v3.
Deterministic first: branch resolution, file/command/error ledgers, masking, budgets, and rendering are Rust. --llm none writes a complete artifact and calls no model.
Optional fold: --llm auto (default), cli:claude / cli:codex / cli:pi, or api:anthropic / api:openai / api:compat/<model>. README says use an API for a long session; a subscription CLI can time out.
Commands: sctxx list, find, extract, expand, verify --strict, doctor, handoff last --to claude.
Skill wrapper: sctxx skill install teaches agents when to call it.

Limitation: 18 stars and 55 crate downloads are early heat, not an audit. The 302 MB to 7.9 KB example is the author's M1 Max synthetic run, not your transcript. CLI backends on a 103,757-event session hit a 600 s timeout in the README; the API path finished in 1 m 37 s. Adding a provider means writing an adapter.

Use Cases

Context-limit handoffs from Claude Code into Codex or Pi without pasting a 300 MB jsonl.
People comparing agents who want a verified, provenance-linked artifact instead of a model-written paragraph.
People who should use ccusage instead if they only needed token cost, not a continuation pack.

Pricing

Piece	Price	Notes from first-party pages 2026-09-13
CLI	$0	Apache-2.0. crates.io 0.3.0. About 18 stars.
`--llm none`	$0 extra	No network.
API fold	Your key	Anthropic, OpenAI, or any OpenAI-compatible base URL.

Getting Started

npm i -g sctxx or cargo install sctxx.
sctxx doctor then sctxx skill install.
sctxx extract claude:last --out .sctxx/ or sctxx extract claude:last --llm none.
Inside the next agent: "use sctxx to extract session 7c1e8f82 from claude code and continue here."

First-party resource: README and docs.

Frequently Asked Questions

Does extract need a model?

No. --llm none is complete. A model is only for semantic judgment.

Can I paste the transcript instead?

That is the problem it claims to avoid. A 302 MB file is neither possible nor useful to paste.

Does it upload logs?

The advertised path reads local stores. Confirm api:* destinations before you fold a private session.

Alternatives

ccusage: usage and cost from local logs, not a handoff.
Claude Code: the session store sctxx reads.
OpenCode: another coding-agent CLI if you wanted to continue there by hand.

Tips

Run sctxx extract --dry-run before spending an API fold.
Prefer api:compat/<model> for long sessions. The README timed out cli:codex at 600 s.
sctxx verify .sctxx/ --strict before you trust a handoff in a new repo.

Conclusion

sctxx is a same-week CLI for moving a coding-agent session across tools without pasting the jsonl. Start at the repo, install, then decide whether --llm none was already enough.