contexto v0.1.11

Contexto is a context engine for AI agents. Instead of summarizing older messages when the context window fills up (which compounds into lossy summaries-of-summaries), Contexto organizes conversation history using hierarchical clustering and retrieves the most relevant episodes via beam search.

Highlights

Hierarchical Clustering with AGNES (@ekai/mindmap)

• Episodes are clustered by semantic similarity using AGNES (Agglomerative Nesting) with average linkage
• Related conversations land in the same branch regardless of when they happened
• Hybrid rebuild strategy: full AGNES rebuild for small trees, O(log N) incremental insertion between rebuilds
• Beam search explores multiple topic branches simultaneously
• Token-budget-aware: fills the context window with the most relevant full episodes, not compressed summaries
• Path tracing for explainability

OpenClaw Context Engine Plugin (@ekai/contexto)

• Sliding-window engine buffers episodes and compacts at a configurable token budget threshold
• Pluggable backend via ContextoBackend interface (ships with RemoteBackend, extensible for custom implementations)
• Default engine delegates compaction to the OpenClaw runtime for drop-in compatibility

What's Changed

• feat: Add contexto mindmap plugin by @DaevMithran in #103
• feat: Migrate to pnpm & update mindmap to contexto by @DaevMithran in #104
• fix: openclaw plugin id by @DaevMithran in #105
• feat: Add @ekai/mindmap by @DaevMithran in #106
• feat: Query via beam search by @DaevMithran in #107
• feat: Add episodic mindmap by @DaevMithran in #109
• fix: Delegate compaction to runtime on default engine by @DaevMithran in #113

New Contributors

• @DaevMithran made their first contribution in #103

Full Changelog: v0-legacy-2026-02-10...v0.1.11