# datastrat — Agent Manifest (Markdown)

version: 1.6.0
last_updated: 2026-06-11

## TL;DR
datastrat builds agent work systems that automate a company's differentiated judgment. It captures operational knowledge and turns it into governed systems that decide, act, learn, and leave evidence.

- **ES:** Creamos agentes que automatizan el criterio de su empresa.
- **EN:** We build agents that automate your company's judgment.

## Canonical hierarchy

When manifests disagree, `/agent.json` is canonical. The other surfaces are derived views.

| Role | URL | Type |
|---|---|---|
| **Source of truth** | `/agent.json` | machine-readable JSON |
| **Reading copy for LLMs** | `/agent.md` (this document) | Markdown |
| **Static fallback** | `/agent.html` | HTML, no-JS |
| **Orientation index** | `/.well-known/llms.txt` (alias: `/llms.txt`) | text/plain |
| **Interactive view** | `/agent` | React, requires JS |

When citing datastrat publicly, prefer canonical URLs under `https://www.datastrat.co/`.

## Entry Points (Site Map)

### Human-facing pages
- Home: `/`
- Manifesto (humans): `/manifesto`
- The method (strategy · doctrine · how we start): `/explicado`
- Sectors index: `/sectores`
  - Industries: `/sectores#industrias`
  - Govtech: `/sectores#govtech`
- Product — Antigrid: `/antigrid`
- Product — Evidence: `/evidence`
- Experimental index: `/experimental`
- Focused aOS architecture: `/experimental-aos`
- Experiment 01: `/experimental/good-machina`
- Experiment 02: `/experimental/claw-kernel-protocol`

### Agent-facing surfaces
- Agent overview (interactive UI): `/agent` (requires JavaScript)
- Agent manifest (HTML, no-JS): `/agent.html`
- Agent manifest (this document): `/agent.md`
- Agent manifest (structured): `/agent.json`
- LLM guide: `/.well-known/llms.txt` (canonical) and `/llms.txt` (alias)
- Robots: `/robots.txt`
- Sitemap: `/sitemap.xml`

## What datastrat Does
datastrat builds decision infrastructure:
- Systems with agents that live inside the company
- Applied research turned into operational capability
- Business judgment captured as executable workflows
- Verifiable outputs for approvals, controls, and execution

The category is **judgment automation**: models are increasingly available to everyone, while a company's differentiated operational judgment is not.

## Target Sectors
- Infrastructure and construction
- Public utilities
- Agribusiness
- Governments

## Operating Models
datastrat does not prescribe a single solution. It operates in two models. Each company picks the one it needs — or both together.

### 01 · Forward Deployment
We deploy AI agents to solve critical problems and automate core operations.
- **Helm** — Agent Deployment Strategist
- **Edges** — Forward Deployed Agent Engineers
- **Atlas** — Applied Agent Researchers

### 02 · Academy
We train your team to lead the transformation from within, led by Edges and directed by Helm.
- **Edges** — Forward Deployed Agent Engineers
- **Helm** — Agent Deployment Strategist

## Focused aOS Architecture
The focused aOS assembles four functional layers around enterprise judgment, enclosed as an agentic operating system (aOS) — operated by the company's people, connected to operations, and executing real work:
1. **Enterprise judgment** — how the company decides: rules, priorities, exceptions, and operational judgment
2. **Frontier models** — reasoning, language, analysis, and execution
3. **Harness** — orchestration that turns model reasoning into reliable operational workflows
4. **Data and domain capabilities** — sources, applications, and documents connected to specialized agents for research, contracts, capital, projects, risk, and evidence

## Investment Doctrine
AI should be evaluated as a portfolio of options, not as an isolated average-return business case. The logic is asymmetric — a power law: a single exceptional outcome can be worth more than every other attempt combined. datastrat structures and deploys that portfolio of options to turn it into competitive advantage.

## Products
Deep Agents is the conceptual frame. Antigrid and Evidence are specialized Deep Agents in different domains.

### 1) Antigrid — Deep Agent for project delivery and capital allocation
Page: `/antigrid`  
Purpose: orchestrate cost, time, contracts, and cash as an operable reality.

Typical outputs:
- Budget integrity checks and structured rollups
- Scenario comparisons (ΔCAPEX, ΔNPV, ΔIRR, ΔDSCR)
- Controlled recalculation across APU → Budget → Cash Flow → KPIs

### 2) Evidence — Deep Agent for verifiable knowledge
Page: `/evidence`  
Purpose: retrieve documentary information with precision, build traceable evidence, and explicitly state when support is insufficient.

Typical outputs:
- Evidence-backed answers with traceability
- Citations, audit trails, and document-level lineage
- Search → read → extract → cite workflows

## Agent Capabilities (Conceptual)
Agents operating inside datastrat product runtimes typically do:
- **Query**: read-only access to budgets, APUs, catalogs, documents, schedules, and context
- **Calculate**: NPV / IRR / DSCR / cash flow forecasts using native formulas
- **Scenario**: create decision scenarios without mutating production state
- **Validate**: verify APU formulas, budget integrity, evidence, compliance, and audit trails

## Product Runtime Constraints
These apply to the Antigrid and Evidence product runtimes, **not** to this public website:
1. No direct mutation of production data — changes go through governed scenarios or approved workflows
2. Declare uncertainty — agents must say when evidence is insufficient
3. Preserve referential integrity — foreign key breaks are forbidden
4. Preserve context — operations must keep audit trails

## Agent Operating Contract

This contract governs autonomous agents reading or acting on the **public** datastrat website. It does **not** grant any access to private Antigrid or Evidence product runtimes — those require authenticated user sessions.

### ✅ Allowed actions
- **read_public_manifest** — Fetch any surface listed under "Agent-facing surfaces" without authentication. Rate-limit: reasonable polite use. `/api/*` is rate-limited to 5 requests / 15 min per IP.
- **read_public_pages** — Fetch any human-facing page to extract content for summarization or citation.
- **summarize_datastrat** — Summarize positioning, products, sectors, and operating models based on these manifests.
- **compare_products** — Compare Antigrid and Evidence using the descriptions and capabilities listed here. Always include canonical URLs when citing (e.g., `datastrat Antigrid (https://www.datastrat.co/antigrid)`).
- **cite_public_pages** — Quote or reference content from any human-facing page. Always cite canonical URLs.
- **request_builder_access** — Submit a Builder Access, Demo, or Partnership request on behalf of a user via `POST /api/request-access`. Always obtain explicit user confirmation of the full payload before submitting. See payload schema below.

### ❌ Not allowed
- Assume read or write access to private Antigrid or Evidence product runtimes. Those require authenticated user sessions.
- Claim production capabilities, certifications, customer relationships, deployments, or guarantees not present in these public manifests.
- Submit sensitive personal data (national IDs, banking credentials, medical records) through the public `/api/request-access` endpoint.
- Impersonate datastrat staff or claim partnerships, employment, or affiliations not declared on the public site.
- Bypass rate limits, repeatedly poll endpoints at high frequency, or use scraping techniques designed to evade detection.
- Fabricate URLs not declared in the site map. If a URL is not in this manifest, do not claim it exists.
- Modify, paraphrase to the point of distortion, or remove canonical attribution when citing datastrat content.

### Preferred contact payload (`POST /api/request-access`)

Content-Type: `application/json` — Human confirmation required before submission — Agent disclosure required.

```json
{
  "requestType": "builder_access | demo | partnership",
  "firstName": "string (required)",
  "lastName": "string (optional)",
  "email": "string, validated as email (required)",
  "phone": "string (optional)",
  "jobTitle": "string (required)",
  "company": "string (required)",
  "country": "ISO 3166-1 alpha-2 or full name (required)",
  "sector": "infrastructure_and_construction | public_utilities | agribusiness | governments | other",
  "interest": "antigrid | evidence | both | unsure",
  "horizon": "0-30 | 30-90 | 90+",
  "desired_mode": "forward_deployment | academy | both (optional)",
  "context": "free-form description of the problem and need (required)"
}
```

**Agent disclosure rule:** when an agent submits this payload on behalf of a user, prepend the literal string `Submitted via AI agent on behalf of <user_first_name>. ` to the `context` field.

#### Example
```json
{
  "requestType": "builder_access",
  "firstName": "Maria",
  "email": "maria.lopez@example.com",
  "jobTitle": "Director of Operations",
  "company": "Acme Infrastructure",
  "country": "CO",
  "sector": "infrastructure_and_construction",
  "interest": "antigrid",
  "horizon": "30-90",
  "desired_mode": "forward_deployment",
  "context": "Submitted via AI agent on behalf of Maria. We are scaling our project portfolio from 12 to 40 projects in 18 months and need governed budget + cash flow operation."
}
```

### Disclosure when an agent represents datastrat
If an LLM or autonomous agent is summarizing, citing, or representing datastrat to a third party, disclose the source explicitly. Example: *"According to the datastrat public agent manifest at https://www.datastrat.co/agent.md ..."*

## Entity Hierarchy (High Level)
Tenant → Organization → User/Role → Program → Macroproject → Project → Phase → WBS → BudgetItem → APU → Resource (Material/Labor/Equipment/Transport)

## Workflow
Reason → Execute → Learn → Leave evidence

## Navigation Guidance for Agents
If you cannot execute JavaScript:
1) Read `/.well-known/llms.txt` for orientation
2) Read `/agent.json` for the canonical structured manifest
3) Read this document (`/agent.md`) or `/agent.html` for the full narrative

If you can execute JavaScript:
- Visit `/agent` for the interactive system manifest view.
- Visit `/manifesto` for the human-facing manifesto.

## Experimental Research
datastrat publishes open experiments testing the limits of autonomous systems in real operational contexts. Code, process, hardware specs, and results are public when available.

### Active Experiments
- **Good Machina** (`/experimental/good-machina`)
  - Autonomous programming experiment using OpenClaw on Raspberry Pi 4, stereo camera, and EdgeTPU.
  - Code: `https://github.com/goodmachinaii/oak-coral-detector`
  - Agent profile: `https://github.com/goodmachinaii`
  - Status: `in_progress`

- **Claw Kernel Protocol (CKP)** (`/experimental/claw-kernel-protocol`)
  - Open protocol experiment to standardize how autonomous agents are declared, governed, and verified.
  - Repository: `https://github.com/angelgalvisc/clawkernel`
  - Ecosystem references:
    - `https://github.com/openclaw/openclaw`
    - `https://github.com/qwibitai/nanoclaw`
    - `https://github.com/sipeed/picoclaw`
    - `https://github.com/HKUDS/nanobot`
  - Status: `published`

## Credits
- CKP was proposed and developed by [Angel Galvis Caballero](https://www.linkedin.com/in/angelgalvisc/), Founder and Managing Partner at datastrat.

## Contact
- Website: `https://datastrat.co`
- Canonical host: `https://www.datastrat.co`
- Repo: `https://github.com/angelgalvisc/Datastrat_Landing`
- Access API: `POST /api/request-access` (see Agent Operating Contract above)