# Butter

> Self-learning context for accurate, verifiable AI workflows. Butter is a business context engine for AI: it connects to the tools your company already uses (Notion, Google Sheets, Slack, Confluence, dbt, Cursor, Claude Code) and maintains a versioned, governed memory of how your business operates. Every LLM query in your stack gets enriched with the right context automatically—so models do less searching, fewer retries, and produce fewer hallucinations.

Canonical site: https://trybutter.ai

Butter is one platform with three product surfaces:

- **Context Management** — a scoped sandbox of company context per team, governed centrally at the data layer.
- **Semantic Memory** — a self-learning, versioned knowledge graph of your business: metrics, definitions, decisions, and SOPs.
- **Inference Cost Management** — a live dashboard of LLM spend by team, query type, and model; cuts token cost through context enrichment, not compression.

Butter runs as an MCP (Model Context Protocol) server. It is model-agnostic and works with Claude, GPT, Gemini, and open-source models.

## Products

- [Products overview](https://trybutter.ai/products): One platform—Context Management, Semantic Memory, and Inference Cost Management. The hub walks through onboarding from install to enriched, auditable prompts in days ([user journey](https://trybutter.ai/products#journey)).

### User journey (Products hub)

From install to enriched prompts in days:

1. **Connect your tools** (~1 hour) — Install Butter, grant access, and enable Butter MCP in Cursor, Claude, or ChatGPT. When your knowledge graph is ready, start calibration.
2. **Calibrate together** (2–5 days) — Butter proposes context files and metric definitions; your team reviews and approves commits so shared knowledge is versioned and governed.
3. **Enrich every prompt** (ongoing) — MCP attaches committed context on every query automatically. Use **Expose context** on any answer to see which nodes, sources, and versions were used.
4. **Context management** (on demand) — Audit the knowledge graph per node: access control, source lineage, and version history—plus token analytics showing spend and estimated savings from enrichment.

Teams can also engage a forward-deployed engineer for onsite implementation: pick use cases, install Butter alongside your team, and hand off when you are self-sufficient.

### Product pages

- [Context Management](https://trybutter.ai/products/context-management): Context management for AI scoped per team. Sandboxes of company context governed at the data layer, with full audit trails.
- [Semantic Memory](https://trybutter.ai/products/semantic-memory): A living, versioned knowledge graph of how your business operates—metrics, definitions, and decisions that sharpen with every query your team runs.
- [Inference Cost Management](https://trybutter.ai/products/inference-cost-management): Live dashboard of AI spend by team, query type, and model. Reduces token waste by enriching prompts up front, not by compressing context away.

## Key concepts

- **Business context engine**: Butter's category—the layer between your company's knowledge and any LLM you use. Distinct from prompt engineering, prompt caching, and traditional semantic layers.
- **Self-learning context**: Context that improves as teams use Butter—grounded in a versioned knowledge graph, not ad-hoc prompt edits.
- **Context governance**: Deciding which slice of company context flows to which team in which AI tool. Butter handles this at the data layer, not the prompt layer.
- **Verifiable AI workflows**: Enriched queries carry lineage—what context was attached, which version, and which sources—so outputs are auditable in production.
- **Versioned knowledge graph**: A living graph of business meaning. Every metric definition, decision, and conflict resolution is preserved with commit time and valid-from / valid-to.
- **Context-tax inference cost**: Most LLM cost comes from models searching for context they don't have, retrying on wrong answers, and generating hallucinated paths. Butter eliminates the context tax up front.

## Pages

- [Home](https://trybutter.ai/): Self-learning context for accurate, verifiable AI workflows
- [Sitemap](https://trybutter.ai/sitemap.xml)