Crate rhema_transcript 

The bounded view over an unbounded session.

Every step sends session.history.clone() (SPEC.md §4): cost grows with the session, and the context window ends the session that outgrows it. This crate splits what was fused. The log — the caller’s Session — stays complete; the view — what a request carries — is a projection built per request under a byte budget. Old messages are represented by the stepped compaction block, nothing is deleted anywhere, and zoom is the model’s way back to any original.

• tree — the slot algebra: message ranges (id, n)
• summary — the extractive line text: heads and counts, no model calls
• compact — the stepped compaction engine: the append-only block, the summarizer seam, the model worker
• view — the request view: the provider decorator, the mirror, the budget, the benchmark stats
• zoom — the zoom tool

The design is specs/rhema-transcript.md.

Re-exports

pub use compact::ModelSummarizer;

pub use compact::PendingSummary;

pub use compact::Section;

pub use compact::SectionOrigin;

pub use compact::Summarizer;

pub use view::LogContent;

pub use view::SkillText;

pub use view::Transcript;

pub use view::TranscriptSettings;

pub use view::TranscriptStats;

pub use view::TranscriptView;

pub use view::UsageBreakdown;

pub use view::WorkerOutcome;

pub use zoom::ZOOM_MAX_BYTES;

pub use zoom::Zoom;

Modules

compact

The stepped compaction engine: what the view says about the aged prefix, and who writes it.

summary

The extractive line text: what the view says about messages it does not carry whole.

tree

The slot algebra: how ranges of the log are named and split.

view

The request view: the provider decorator that rewrites req.messages under a byte budget, and the mirror zoom reads.

zoom

The zoom tool: the model’s way back to any original message.

Functions

attach

Wires a transcript over inner: the view decorator plus the zoom resolver, sharing one mirror. A consumer that builds its own rhema_core::Agent sends the wrapped provider through to the builder and overlays the resolver over its tool chain; the coding agent does both in its constructor when its config enables the transcript.

attach_with_summarizer

attach with the compaction’s model fill: summarizer rewrites each new section’s extractive stand-in into a dense block at flush time. None is the extractive-only compaction. Returns the mirror too, for a driver that reports the benchmark stats.