# PrismLib

One canonical identity. Human page: https://www.insightits.com/products/prismlib.html

| Field | Value |
|-------|-------|
| Slug | `prismlib` |
| Status | published |
| Family | supporting |
| Infra category | — |
| Record type | product |
| Parent | — |
| JSON | https://www.insightits.com/catalog/prismlib.json |
| Markdown | https://www.insightits.com/catalog/prismlib.md |
| GitHub | https://github.com/insightitsGit/prismlibplusapi |
| PyPI | https://pypi.org/project/prismlib-plus/0.7.0/ |
| Install | `pip install "prismlib-plus[enterprise,cache,fabric]"` |
| Version | 0.7.0 |

## What it is

Open-source library with PrismCache (in-process semantic LLM cache) and PrismDriver (WAL-streamed DB driver). Full stack ships as prismlib-plus. ChorusMesh is the paid cluster companion on the same landing — different from CHORUS Fabric and ChorusControl.

## Problem

Repeat LLM calls and 100ms+ cloud DB reads without an in-process cache or WAL-streamed local index.

## Alternatives

- **GPTCache / Redis semantic cache** — Product landing compares hit-rate and infra (PrismCache in-process vs Redis+FAISS). Vendor-authored Azure westus2 measurements; not a third-party bake-off page.

## Features

- **PrismCache — 95.9% Hit Rate** — Semantic paraphrase matching in-process. No Redis, no Pinecone. Mathematical multi-tenant isolation via JL projection seeded by SHA-256(tenant_id).
- **PrismDriver — 439× Faster Reads** — WAL/binlog streamed via CHORUS Fabric to a local PrismResonance index on the app node — 0.27ms reads vs 118.5ms baseline (Azure e2e).
- **PrismAPI — 83% Fewer Embeds** — Vector-native agent API: provider embeds once, consumers receive pre-projected float32 over CHORUS — no re-embedding on retrieval.

## Architecture

Local PrismCache → cluster cache (TOKEN_SYNC) → compression → dedup → LLM call. PrismDriver uses a prism-wrapper daemon and WAL invalidation. Optional CHORUS fabric transport in prismlib-plus[fabric].

## Use cases

- SaaS embedding-bill reduction (PrismAPI)
- Local DB reads vs cloud round-trips (PrismDriver)
- Multi-replica FAQ bots (cluster cache)

## Benchmarks

Azure westus2 product measurements (1 vCPU / 2 GiB): PrismCache mixed hit rate 95.9% (50 users × 300s); PrismDriver 439× vs 118.5ms cloud reads; PrismAPI 83% fewer embed calls at top_k=5.

### Disclosures

- Vendor-authored Azure e2e measurements. Not a third-party bake-off.

### Evidence URLs

- https://github.com/insightitsGit/prismlibplusapi/blob/main/BENCHMARK_RESULTS.md


## GitHub

https://github.com/insightitsGit/prismlibplusapi

## PyPI

https://pypi.org/project/prismlib-plus/0.7.0/

## Documentation

- [Release notes](https://github.com/insightitsGit/prismlibplusapi/blob/main/RELEASE_NOTES.md)
- [Benchmark results](https://github.com/insightitsGit/prismlibplusapi/blob/main/BENCHMARK_RESULTS.md)
- [Enterprise docs](https://github.com/insightitsGit/prismlibplusapi/blob/main/ENTERPRISE.md)
- [Whitepaper PDF](https://www.insightits.com/docs/prismlib-whitepaper.pdf)
- [githubBase](https://github.com/insightitsGit/prismlib)
- [issues](https://github.com/insightitsGit/prismlibplusapi/issues)
- [githubBaseIssues](https://github.com/insightitsGit/prismlib/issues)
- [pypiBase](https://pypi.org/project/prismlib/0.4.0/)
- [prismApiDocs](https://github.com/insightitsGit/prismlibplusapi/blob/main/PrismAPI.md)
- [prismresonanceGithub](https://github.com/insightitsGit/prismresonance)
- [chorusFabricGithub](https://github.com/insightitsGit/chorus-fabric)

## Is not

- CHORUS Fabric
- ChorusControl
- A hosted SaaS cache

## Related products

- [python-libs](./python-libs.md)
- [chorus-fabric](./chorus-fabric.md)
- [choruscontrol](./choruscontrol.md)