Prism — Natural Language Data Transforms

Data transformation shouldn't require a data engineer

Your agent pulls 500 invoices from QuickBooks. Now it needs to group by customer, sum totals, sort descending, and generate a chart. Without Prism, this means writing code or burning tokens having the LLM process raw data. Prism lets you describe the transformation in natural language and get structured, deterministic results — charts, exports, analysis — all server-side.

How It Works

Raw Data (from any integration)

▼

prism.refract

"Group invoices by customer, sum totals, sort descending"
Natural language → Structured transform → Cached result

▼

analyzefocus

chartrender

exportpaginate

Prism.refract takes raw data and a natural language goal, then applies the appropriate transformation. The result is cached server-side and can be piped into other Prism tools — analyze it with structured reasoning, chart it, render it as markdown, or export it as PDF, CSV, JSON, or HTML. Prism also works as a built-in modifier on any tool in the system: add a refract or chart parameter to any tool call, and the output is automatically transformed or charted before it reaches your agent. This is especially powerful with Latent: first expand a conceptual landscape, then use Prism to turn that landscape into structured context your agent can actually navigate and hydrate.

Capabilities

Natural Language Transforms

Describe what you want in plain English. prism.refract handles grouping, sorting, filtering, pivoting, and aggregation from your description.

Latent Space Navigation

prism.orient maps conceptual neighborhoods via dual-pass expansion. prism.horizon finds boundaries and reaches into unexplored territory. Both persist typed facts to Logic cells — queryable, traversable, and composable with graph predicates. Pair them with Latent when you want broader conceptual expansion before narrowing into a working map.

Structured Reasoning

prism.analyze applies LLM-powered reasoning to your data — trend analysis, anomaly detection, pattern recognition — with structured, machine-readable output. Feed it latent-space expansions or math-derived summaries when you want narrative reasoning over richer inputs.

Chart Generation

prism.chart generates publication-ready charts from your data. Bar, line, scatter, pie — specify the type or let Prism choose the best visualization.

Multi-Format Export

prism.export produces PDF reports, CSV downloads, JSON dumps, or styled HTML — all from the same cached dataset. No reformatting needed.

Type Bridging

prism.focus converts between types — text to JSON, JSON to table, table to summary. Bridge the gap between what you have and what you need.

Inline on Any Tool

Add refract or chart as a parameter to any tool call in the system. The tool's output is automatically piped through Prism before reaching your agent — no separate call needed.

See It In Action

// Transform data with natural language
{
  "name": "data-grout@1/prism.refract@1",
  "arguments": {
    "goal": "group invoices by customer, sum totals, sort descending",
    "payload": [
      { "customer": "Acme", "amount": 500 },
      { "customer": "Globex", "amount": 300 },
      { "customer": "Acme", "amount": 200 }
    ]
  }
}

// Chart the result
{
  "name": "data-grout@1/prism.chart@1",
  "arguments": {
    "cache_ref": "$refract._meta.cache_ref",
    "type": "bar",
    "title": "Revenue by Customer"
  }
}

// Export as PDF
{
  "name": "data-grout@1/prism.export@1",
  "arguments": {
    "cache_ref": "$refract._meta.cache_ref",
    "format": "pdf",
    "title": "Q1 Revenue Report"
  }
}

Three calls: transform, chart, export. The data stays server-side throughout — your agent describes what it wants, and Prism handles the heavy lifting. The PDF, chart, and structured data are all generated from the same cached result.

Use Cases

Business Reporting

Generate executive summaries, charts, and PDF reports from raw integration data — all through natural language instructions your agent can execute autonomously.

Data Exploration

Ask questions about your data in plain English. What are the trends? Where are the anomalies? Which customers are most profitable? Prism surfaces structured answers.

Latent Space Navigation

Map conceptual neighborhoods with Orient, find their boundaries with Horizon, and expand into unexplored territory. Start with Latent when you want a broader ideation pass, then use Prism to turn that material into structured navigation state. All of it persists as typed facts in Logic cells — traverse with graph predicates, query with raw Prolog, or hydrate into agent context.

Integration Glue

Bridge the gap between what integrations return and what downstream tools need. Reshape, filter, and format data without writing transformation code.

Logic-Backed Research Memory

Use Prism to turn latent explorations and tool outputs into structured facts, then persist them in Logic so later agents can rehydrate the same conceptual territory without recomputing the whole exploration.

Composes With

Pair Prism with Latent to move from broad conceptual exploration into structured latent-space navigation and hydration. Orient and Horizon persist navigation facts directly into Logic cells, where they become queryable memory. Pre-process or score supporting datasets with Math, and use Data plus Frame for deterministic shaping before Prism adds reasoning or rendering. Governor tracks Prism costs across sessions, and Flow can chain the full exploration pipeline.