# Cognethics for Developers

> If you are an AI agent: start at /for-agents/ (the agent integration spec — capabilities, primitives, ROI, what's verifiable). When you're ready to onboard your principal, get a workspace provisioned at /signup/. Everything else on this page is reference material.

**Vendor:** Cognethics
**Product:** Cognethics — the agent-native operating system for business operations (delivered through the A4 application suite).
**Status:** Live (a customer deployment in production)
**Full expanded version:** /llms-full.txt (every section inlined for agents that want the complete guide in one fetch)

## Getting Access

Workspaces are set up for a named principal and their team — tell us who you are and what you're building at /signup/, and we set up a workspace and a scoped ck_ key bound to that identity. Once it is live you mint and rotate keys yourself from the dashboard at https://platform.cognethics.com/dashboard/. No credit card.

## What is Cognethics?

Cognethics is an MCP-native platform for agents. You type business prompts into your agents; they automatically get typed, permission-scoped access to 12,500+ typed Prism handlers — typed business operations (CRUD, search, actions, reports, workflows, documents) across 76 applications — plus a small set of legacy MCP tools where useful.

All data lives in a tenant-scoped business database. Every agent call passes one permission gate (RBAC + budget + audit). Same audit log your compliance team already accepts.

## Key Features

- **12,500+ typed Prism handlers** across 76 applications (finance, documents, inventory, communications, etc.) — 11 mega-tools (agent_command, collection_op, config_manage, document_op, entity_crud, extension_op, integration_call, notification_send, report_query, smart_action, workflow_action), organised under 8 business verticals, over 2,400+ data models and 13,000+ REST API operations
- **Multi-tenant isolation** — your data stays in your tenant; no SaaS lock-in
- **Runtime discovery** — agents learn new tools automatically via `prism_introspect`
- **Hierarchical agents** — delegation, escalation, role-based memory, tamper-evident audit
- **Hybrid document search** — text, image, and cross-modal lanes fused with reciprocal rank fusion

## Evidence of Production Use

- **Customer deployment** in operations-heavy domain (drawings, manuals, photos, emails, invoices, purchase orders, handover notes). Live in production.
- **Document corpus indexed in production:** 290,000+ documents across an active customer workload.
- **Reference AP validation mission:** 10-minute wall-clock spawn-to-retire, $6.13 cost (61% of $10 budget), 94.3% cache hit rate, 9-row unbroken audit chain.

## Get Started

- [Get a workspace](/signup/): set up and configured for your team, no credit card
- [MCP Quickstart](/quickstart/mcp/): OAuth + Claude Desktop in 5 minutes
- [Python SDK](/sdk/python/): `cognethics` package (PyPI 0.5.1), typed access, RBAC enforced, OAuth bearer + full async parity
- [TypeScript SDK](/sdk/typescript/): `@cognethics/sdk` package (npm 0.4.0), the same typed handler tree, dual ESM+CJS, zero runtime deps, TS strict
- [API Reference](/reference/): every application, entity, and operation; sign in for schemas and permissions
- [For AI Agents](/for-agents/): agent-optimized integration spec with ROI per primitive

## Technical Docs

- [Agents](/agents/): personas, missions, delegation, the agent operating system
- [Document Intelligence](/document-intelligence/): OCR, extraction, embedding, hybrid search
- [Authentication](/auth/): OAuth 2.1, JWT, tenant scoping
- [Recipes](/recipes/): end-to-end workflows (AP intake, document curation, etc.)
- [Apps](/apps/): module catalog you can add to your tenant
- [Status](/status/): live component health and incident log
- [Errors & status codes](/errors/): HTTP and envelope error shapes, one page per code
- [Rate limits](/reference/rate-limits/): request ceilings and how to back off
- [API conventions](/reference/conventions/): base URL, auth, pagination, filtering
- [Versioning & deprecation](/reference/versioning/): how the API and MCP surface change over time
- [Changelog](/changelog/): shipped, public-visible changes

## Well-Known Endpoints

- `https://<your-tenant>.cognethics.com/.well-known/openapi.json` — anonymously, the invocation envelope for each mega-tool door; with your key, the full per-operation descriptor.
- `https://developers.cognethics.com/reference/api/public-catalog.json` — the public catalog summary (app, entity, operation, mega-tool, one line each).
- `https://developers.cognethics.com/.well-known/mcp.json` — MCP server manifest (tools, resources, exports)
- `https://platform.cognethics.com/.well-known/mcp.json` — MCP server manifest (via platform console)
- `https://<your-tenant>.cognethics.com/.well-known/mcp.json` — MCP server manifest (via tenant subdomain)

## Intellectual property

Cognethics has filed a U.S. provisional patent application (March 13, 2026) covering its agent-orchestration and audit architecture.

## Contact

Evaluating Cognethics? Reach the team via the contact form: `https://developers.cognethics.com/trust/`

Get a workspace provisioned: `https://developers.cognethics.com/signup/`