Tesserae v0.10.0 — interop, fast transcript search, and machine-wide setup

<!-- translations:start -->

한국어 · 中文 · 日本語 · Русский · Español · Français · Deutsch

<!-- translations:end -->

Released 2026-06-20 · PyPI · GitHub release · pip install --upgrade tesserae==0.10.0

A feature release: knowledge interchange, faster session search, and one-shot machine-wide setup. Everything is additive — a plain tesserae compile is unchanged, and per-project config still overrides the new global defaults.

1. Google OKF v0.1 import/export

tesserae export okf writes the graph as a Google **Open Knowledge Format** v0.1 bundle — a tree of Markdown files with YAML frontmatter and relative-link relations. --import DIR reads one back. Tesserae's own bundles round-trip losslessly via an x_tesserae frontmatter namespace (real node ids + typed edges + edge metadata); foreign bundles import best-effort (unknown type → Concept, untyped links → references), tolerating missing types and broken/symlinked files per the spec.

2. Fast transcript search on the sessions dashboard

tesserae serve's sessions dashboard gains a full-text search box over every indexed Claude/Codex transcript, powered by MD1 (BM25). It's optional and graceful: without the memex binary the box shows an actionable message and the rest of the dashboard is unaffected. The GET /api/transcript-search endpoint is gated to same-origin/loopback callers so a visited web page can't probe local history, and results render via textContent (no markup injection).

3. memex-style read-discipline for MCP

Inspired by Databricks' "programmable scratchpad," compile_context now accepts preview=N: instead of dumping a huge body into the agent's context, it returns a bounded preview plus a content-keyed handle, and the new get_handle tool pages the rest on demand. Default preview=0 keeps the full-body behavior.

4. Extraction quality signals

Session findings now carry confidence + confidence_rationale + revisit_signals distilled from the extractor (ideas from Jonasb8/memex; no code copied). They ride the content-keyed extraction cache, so they stay byte-idempotent, and surface in the fresh_insights MCP tool as flags (never truth guarantees).

5. Machine-wide setup + dependency installs

tesserae config setup sets LLM defaults for all projects (--llm-provider, --reasoning-effort, --codex-home, --claude-config-dir) and installs optional dependencies (--install all) in one shot. tesserae config deps lists/installs memex, cognee, raganything, and understand-anything; tesserae config llm gained --reasoning-effort; and tesserae init offers to install memex during an interactive setup. Resolution order: env → project config → machine-wide config → built-in.

Upgrading from v0.9.x

Drop-in. Run tesserae config setup to configure the machine-wide defaults once, or keep your per-project .tesserae/config.json — it still wins.