Context Capsules

The usual agent loop is search, read, search again, read more, then guess when the context window runs out. A Capsule replaces that with one compiled artifact: the passages that actually answer the goal, under a single token budget, each carrying its exact line span and content hash, plus a written statement of what could not be found. Your agent reads it once. You can verify it months later.

Install GNO Read the docs

Use cases: Citation-complete agent handoffs; Multi-document comparisons and decision support; Repeatable evidence packets for audits and reviews; Long-running work where the sources move under you

What it gives you

One global token and byte budget across every selected passage
Exact source line ranges, hashes, dates, headings, and collection identity
Diverse evidence selected by marginal facet coverage
Overlapping passages collapsed and omitted candidates explained
Explicit evidence gaps instead of invented synthesis
Deterministic JSON plus readable Markdown projections
Non-mutating verification for unchanged, stale, missing, and reranked evidence
48.94% fewer retrieval calls and 44.12% less model-visible context across 48 paired benchmark tasks, with task completion accuracy unchanged
Available identically from the CLI, SDK, REST, and MCP

Try it yourself

Representative commands and entry points. Full reference lives in the documentation.

gno context build 'compare the proposals' --collection work-docs --budget 12000 --json --output capsule.json

gno context verify capsule.json

gno context watch capsule.json --notify

gno context reverify <registration>

Keep reading

Related features and docs.

Context CLI reference

/docs/cli

Open

Knowledge Delta

/features/knowledge-delta

Open

Verified Ask

/features/local-llm

Open

All features

Back to the feature index

Browse all