AI Data Platform

Data infrastructure
for serious teams.

A unified platform for AI-powered data products, network intelligence, and document understanding — deployable on your terms, in your environment.

Capabilities

Five integrated modules.

Each module is independently licensable. Run what you need, ignore what you don't.

💬
Chat

Chat & Projects

AI-native workspaces for structured research. Conversations that generate artifacts — not just text. Bring your own LLM backend.

🗄️
Headwaters

Data Products

Connect, query, and publish governed data products across heterogeneous sources. DuckDB-powered analytics with a SQL-first interface.

🌐
Wyatt

Networks & Interchange

Decentralized peer-to-peer data interchange with a cryptographic audit ledger. Build ontology-defined knowledge networks across orgs.

Vista

Data Apps

First-class data application runtime. Deploy AI-driven analytical workflows as purpose-built applications for business users.

📄
Doc

Document Intelligence

Extract, index, and reason over unstructured documents. Semantic search powered by Qdrant vector storage, structured output via LLM pipelines.

🔗
Platform

AI Agent Framework

Router → Planner → Executor pipeline. Provider-neutral: run Anthropic Claude or Google Gemini. Swap backends without rewriting prompts.

How it works

The platform, explained.

Not a wrapper around an LLM API. A structured system with real layers.

01

A provider-neutral agent pipeline

The AI engine uses a three-phase pipeline: Router classifies intent, Planner proposes a structured plan that the user confirms, Executor generates the final artifact. Each phase is decoupled from the underlying model. Swap LLM providers — the prompts, the routing logic, and the output schema don't change.

This means you're not locked to a model vendor's roadmap. When a better model ships, you swap a config value, not a codebase.

02

Federated query, not data movement

Headwaters runs DuckDB in-process against your sources. Queries execute close to the data — no ETL pipeline, no warehouse copy, no replication lag. Data products are governed query views: versioned, permissioned, and composable. The source stays where it is.

When a user asks a question, the answer is computed from live source data, not a stale snapshot loaded into a third system.

03

Decentralized interchange with a cryptographic ledger

Wyatt's peer network has no central broker. Organizations connect directly, governed by a shared ontology. Every data exchange is recorded on a cryptographic audit ledger — tamper-evident by construction, not by policy.

This lets regulated industries share data across org boundaries without trusting a third-party intermediary or standing up shared infrastructure.

04

One platform, not twelve integrations

Most teams building AI-powered data tooling end up with a chat tool, a BI layer, a data catalog, a network graph, and a document store — each with its own auth, its own user model, and its own ops burden. Huckleberry is one deployment: one auth layer, one admin surface, one license, one schema for all five surfaces.

When a user in Chat references a Data Product, that link is real — not a screenshot embedded in a message.

Deployment Models

Your infrastructure.
Your rules.

Three deployment postures to match your security requirements, compliance needs, and operational preferences.

Airgapped

Fully Offline

Zero external calls. Cryptographically signed licenses, tamper-evident audit ledger, no usage telemetry. Runs in disconnected environments — SCIFs, regulated infrastructure, or private clouds with strict egress controls.

  • No network egress required
  • Cryptographic license validation
  • Immutable audit trail, local only
  • Flat annual license, no metering
Ideal for Defense, intelligence, financial compliance, air-gapped research environments.
Managed

Huckleberry Cloud

Fully managed, hosted by Didati. No infrastructure to provision, no patches to apply. Metered by consumption — users, API calls, data volume. Get started in hours.

  • Zero ops burden
  • Consumption-based pricing
  • Automatic upgrades
  • SLA-backed uptime
Ideal for Teams that want to move fast without a dedicated infrastructure team.
Infrastructure

Built on the right primitives.

No proprietary lock-in at the data layer. Standard open-source engines, well-understood operational profiles.

DuckDB

In-process analytical engine. Columnar query performance without a data warehouse. Powers all data product queries and cross-source joins.

PostgreSQL / SQLite

Relational backbone. PostgreSQL for production multi-tenant deployments, SQLite for zero-config local or single-tenant installs.

Qdrant

Optional vector store for semantic search and document intelligence. Run alongside the platform or connect an existing instance.

Redis

Agent state persistence. Streaming AI agent sessions are durable across reconnects — no state loss on network interruptions.

LLM Backends

Connect any supported LLM provider using your own API key. No Huckleberry intermediary in the AI call path — your credentials, your rate limits, your spend.

Provider-Neutral Agents

The agent pipeline decouples routing, planning, and execution from the underlying model. Swap LLM providers without rewriting prompts or changing agent logic.

Enterprise

Governance from day one.

Administration features that belong in an enterprise platform, not bolted on after the fact.

SSO / Identity Providers

SAML and OIDC integration. Bring your existing IdP. User provisioning flows from your identity layer, not Huckleberry's.

Role-Based Access Control

Scope-based permissions on every resource. Granular read/write/admin separation at the module, object, and operation level.

Group Management

Org-level groups mapped to permission sets. Manage access at scale without per-user configuration.

Multi-Factor Authentication

Built-in MFA via email and SMS. Enforceable at the org level. No third-party authenticator integration required.

API Keys

Programmatic access with scoped API keys. Machine-to-machine authentication without session management.

Audit Log

Tamper-evident, queryable audit trail for all administrative and data-access events. Exportable for SIEM ingestion.

Whitelabel Branding

Replace Huckleberry's UI chrome with your organization's brand. Full color, logo, and domain customization.

Third-Party Integrations

OAuth-based developer app registration. Extend the platform through a governed integration surface with scoped access.

Cryptographic License Validation

Licenses are signed artifacts. Validation is local and deterministic — no license server dependency in the critical path.

Get started

Ready to talk?

We sell direct. No SDR sequences, no qualification forms. If you're evaluating, email us.