PRODUCT: FOUNDRY

Don't just compile documents. Manipulate them.

The missing actuation layer for AI. STRUC Foundry is a C++ headless engine that exposes a direct object graph to write, edit, and mutate high-fidelity documents via atomic transactions. Engineered for <50ms latency.

* Benchmark: Native C++ Engine. Zero-serialization overhead. No headless browser.
// Direct Object Graph Injection (JSON-RPC)
const response = await fetch('https://api.struc.so/foundry/mutate', {
  method: 'POST',
  body: JSON.stringify({
    operation: 'InjectBlock',
    target: '#compliance_module',
    payload: {
      type: 'policy_block',
      content: 'Data Retention Policy: 5 Years...',
      attributes: { style: 'iso_standard' }
    }
  })
});

// The engine returns the new revision hash instantly
const { revision_id, render_url } = await response.json();
THE END OF SLOW PYTHON WRAPPERS
Python/XML Parsing
2400ms
Cloud APIs
1200ms
STRUC Foundry
42ms

> We solved the physics problem. Document state lives in the hot path of memory, eliminating the 2-second serialization penalty.

PRODUCT: INDEX

STRUC Index: Deep Structure Extraction 🧠

Turn Documents into Data at Scale. Deterministic C++ parsing for speed, with Probabilistic AI offload for high-entropy input. Structure extraction at never-before-seen scale.

LEGACY TEXT SPLITTING
# Chunk 104 (500 chars)
...shall indemnify the Provider against
any claims arising from use of the

# Chunk 105 (500 chars)
Software. Furthermore, the limit of
liability shall not exceed...

❌ Context Shredding: Turns your RAG into an expensive, dumb search engine.

STRUC INDEX EXTRACTION
// Extracted Entity: Financial Table (Q3)
{
  "id": "table_q3_rev",
  "type": "financial_table",
  "data": { "revenue": 50000, "yoy": "+12%" },
  "context_path": "Report > Financials > Q3"
}

✅ Full Object Fidelity. Repairs Broken Tables.

PRODUCT: REFINERY

STRUC Refinery: The Entropy Filter 🛡️

The Refinery sits between your LLM and the Engine. It purifies probabilistic intent into deterministic execution. Zero hallucinations allowed.

REFINERY::GATEWAY IDLE

1. Ingest Intent

Point your agent (LangChain, AutoGen) to the Refinery. We accept raw, messy tool calls from GPT-4 or Claude 3.5.

2. Refine & Validate

The Refinery checks every parameter against the strict Document Schema. Hallucinated attributes are rejected before they touch the state.

3. Deterministic Execution

Once purified, the command hits the Foundry Engine. Execution is atomic, <50ms, and thread-safe.

Supported Compliance Environments

> LegalTech Deterministic redlining. Strict version control.
> FinTech High-volume automated reporting. Audit trails.
> Healthcare Tamper-proof, audit-ready patient records.
> GovTech ISO-compliant archival formats for long-term storage.
MANIFESTO // WHY NOW

The Chatbot Era is Stalling.
Stop chatting. Start compiling.

We have trapped enterprise intelligence inside ephemeral chat windows. We dump structured data into an LLM, get a brilliant answer, and then... it disappears.

The next phase of Enterprise AI isn't about better conversation. It’s about Automated Operational Artifacts.

01. Data as Input

Pipe live Q3 sales figures or clinical trial results directly into your secure LLM. Treat data as a raw stream, not a static file.

02. Agent as Logic

The model analyzes the data and determines the narrative. It doesn't write the document; it plans the mutation.

03. STRUC as Compiler

Instead of hallucinating a file, the agent targets a strict schema. We enforce structure (tables, headers) at the object level. If it breaks the template, the build fails.

04. Deterministic Outcome

The result is a finalized, immutable system-of-record (PDF, DOCX), generated in milliseconds. Generative Operations, not Generative Text.

Direct Object Access

Stop treating documents like flat files. Interact with a persistent C++ object graph. Inject rows, update variables, and reorder sections via RPCs without overhead.

Strict Schema Enforcement

LLMs hallucinate formatting. We don't. Our C++ engine enforces rigid templates at the object level. The AI suggests changes; the Engine guarantees validity.

Deep Structure Extraction

Deterministic Speed + Probabilistic Intelligence. Our C++ parser handles 95% of structure instantly. For the complex 5%, we seamlessly offload to vision models. Scale without compromise.

High-Fidelity Roundtripping

Cycle documents from DOCX -> Object -> DOCX. Our engine preserves the full semantic state—comments, track changes, and margins—so you can automate workflows safely.

State Reconciliation

Never lose a redline. Our engine tracks every atomic mutation, resolving conflicts and preserving revision history across concurrent AI and human edits.

Universal Output Targets

Engineered for Paged Media. Compile Object -> DOCX / PDF from a single source of truth. Ensure your web preview matches the final document pixel-for-pixel.

Request API Keys

We are currently in Private Beta. We are manually onboarding high-volume partners to ensure environment stability.

/ to focus

THE BARRIER

Magic vs. Utility.

It is easy to look at a fluid conversation with ChatGPT and think the enterprise is about to be disrupted overnight. But there is a massive gap between Generating Cool Stuff and Executing Business Processes.

  • A legal contract cannot have "mostly" correct cross-references.
  • A financial report cannot have "hallucinated" table rows.
  • A compliance audit cannot have "approximate" version history.

In the enterprise, "95% correct" is not an optimization. It is a failure state.

"We are getting distracted by the chat interface. Real work doesn't happen in a chat window. It happens in documents, workflows, and systems of record.

The disruption won't come from a smarter chatbot. It will come from the infrastructure that allows AI to adhere to strict working practices."
```