Skip to main content
DeepMyst is built like infrastructure. Six layers, where every request inherits identity from the bottom and policy from above. The same stack that runs your first agent runs every agent — there’s no second system for production.

​
The six layers

LayerWhat it isIncludes
ShellHow humans use itChat, copilot, embeds, public portals
AgentsFirst-class processesVoice, canvas, document, workflow, kanban, form, and more
MemoryWhat agents knowKnowledge graph, documents, datasources, history
SkillsWhat agents can doBuilt-in tools, MCP, web search, sub-agents
ModelsWhat powers their thinking500+ models, OpenAI-compatible, auto-failover
GovernanceThe rules everything inheritsSSO, RBAC, cascading policy, KYT, audit

Shell

The surfaces people touch. Embed an agent anywhere, share a link, or hand off to a chat surface.

Agents

First-class processes in a dozen shapes — voice, canvas, document, workflow, kanban, form, and more.

Memory

The persistent layer every agent reads from: knowledge graph, documents, connected datasources, and history.

Skills

The tools and integrations every agent can invoke — built-in tools, MCP, web search, and sub-agents.

Models

One API across every leading model. Swap providers without rewriting code, with automatic failover.

Governance

Identity, cascading policy, inline security, and audit — inherited by every request, logged forever.

​
How a request flows

Every call — whether from an agent or a direct API request — runs the same pipeline. DeepMyst handles identity, policy, security, optimization, and metering around a pluggable gateway hop:
  1. Auth — validate the API key
  2. Policy lookup — resolve the key → app → dept → org cascade
  3. KYT inline classification — injection, PII, toxicity, and intent in under 15ms
  4. Guardrail enforcement — block, flag, or redact based on policy
  5. Credit & rate check — enforce RPM/TPM limits and budgets
  6. Optimization — losslessly compress context
  7. Gateway — route, proxy, and return from the chosen model
  8. Metering — record usage and calculate cost
  9. Enrichment — async KYT Pass 2 and behavioral analytics
  10. Events — emit webhooks, write the audit log, update analytics
The model provider never sees user identity, classification results, or uncompressed requests — DeepMyst handles those before and after the gateway hop.

​
One API, every model

The Models layer is an OpenAI-compatible gateway. Point your base URL at DeepMyst and the same SDK works across 500+ models from every major provider. A/B test one model against another with a single config flip — your code never changes. 
from openai import OpenAI

client = OpenAI(
    api_key="your-deepmyst-api-key",
    base_url="https://api.deepmyst.com/v1",
)

# Swap the model string; everything else stays the same
client.chat.completions.create(
    model="claude-opus-4-5",   # or gpt-5, gemini-3-pro-preview, ...
    messages=[...],
)

​
Pluggable gateway

The Models layer is designed to run behind multiple gateway adapters through a single interface, so routing infrastructure can change without touching the layers above it. The key principle: the gateway adapter never sees user identity, classification results, or uncompressed requests — those are handled by DeepMyst core, so optimization and governance stay consistent no matter what’s underneath.

​
Next steps

Agent Builder

Build on the Agents layer in human language

Governance

Configure the rules every layer inherits