Measure
Record original and compiled bytes, model-neutral estimated tokens, reduction percentage, source coverage, and duplicates removed.
[ Maha Strategies // epistemic middleware ]
Give agents less context. Keep the evidence path.
The Maha Context Compiler turns a bounded set of raw documents into a task-specific Context Pack: deduplicated passages, source references, hashes, and transparent before/after measurements. It is deterministic middleware, not a model wrapper.
Trace
Every included passage points back to a source ID, passage position, and content hash. The original document is never stored.
Constrain
Set a context budget and an explicit task. Passages that do not fit are excluded visibly rather than silently overflowing the prompt.
What this first release does not claim
It does not promise a fixed percentage reduction, exact provider token counts, complete source coverage, factual verification, or hallucination prevention. Those are workload-specific outcomes to measure. Use MPS when consequential claims need a separate provenance review.
Machine integration
Authenticated clients call the Context Pack API with a task, a token budget, and 1–8 textual sources. The response is returned directly; the service retains only a privacy-safe outcome ledger, not source text or the compiled pack.
How it unifies Maha
Context Packs are the portable evidence object between the corporate systems: MPS can review their consequential claims; the Enterprise MCP Gateway can restrict which tools receive them; and the commercial ledger can measure where compact, source-linked context is actually used.