v1.1-ace-live · Phase C · 92-Layer Architecture

Build on the A380

MCP Federation · FOSS Stack · 92-Layer Architecture · 75 Constitutional Laws · 65 AI Agents

Get Started API Reference GET /context/orchestrai ↗

Quick Start

🔌

Connect via MCP

MCP federation endpoint setup — connect any MCP-compatible client to the OrchestrAI convo-bridge.

# convo-bridge endpoint
convo.orchestraios.com/mcp

🤖

Deploy an Agent

Agent lifecycle, IAT tier assignment, CVO registration. Every agent follows the ADLC triple-gate.

# CVO register
INSERT INTO cvo_agents
(id, name, iat_tier...)

🔭

Glass Dome API

Access full OS state, layer list, agent status, and live metrics via ctx.orchestraios.com.

GET /context/orchestrai
GET /context/layers
GET /health

📊

Run the Eval Suite

7-tier evaluation framework — from unit tests to adversarial red-team and IndiaFinBench domain eval.

# Run EV-1 unit tests
pytest tests/ -v
# Run EV-8 promptfoo
promptfoo eval

Architecture Reference

OrchestrAI OS is organized into 5 planes with 92 CVO-registered layers, 65 agents across 7 IAT tiers, and 75 constitutional laws.

Layer Reference — 5 Planes, 92 Layers

Plane	Layer Range	Count	Purpose	Key Layers
D — Data	L1–L18	89	Data ingestion, storage, streaming, observability	SL1 · SL3 · SL16 · SL-MEMDIR
S — Service	L19–L45	89	Agent orchestration, MCP, LLM routing, auth	SL-MCPAG · SL-CPOD · L-FSI-01
B — Business	L46–L60	6	Domain logic, compliance, financial workflows	FinMesh layers · KYC/AML · GL
O — Operations	L61–L75	6	CI/CD, deployment, infra automation, watchdog	AGT-INFRAWDG · ADLC-gates
L — Live	Live flag	25+	Layers promoted to live=true in CVO — operational in production	All layers with ADLC=live

Agent Reference — 7 IAT Tiers, 65 Agents

IAT Tier	Maturity	Definition	Example Agents
IAT-1	Operational	Live in production, CVO-registered, 25+ tests passing	AGT-ACELF01 · AGT-INFRAWDG · Aria
IAT-2	Integrated	Built, wired to OS, CI passing, not yet production traffic	AGT-FV01 · AGT-RSCAN · AGT-MCPAG
IAT-3	Tested	Built, unit tests passing, not yet wired to OS bus	Constitutional Pod agents (9)
IAT-4	Scaffolded	Spec complete, skeleton built, tests pending	Memory Director sub-agents
IAT-5	Spec-only	CVO-registered, spec exists, no code yet	FSI Advisory extended suite

Law Reference — 75 Constitutional Laws

GOVERNANCE

LAW-0 ADLC triple-gate (verify→plan→execute)

LAW-10 Verify-don't-relay

LAW-14 SoT from CVO DB only

LAW-16 No fabricated IDs

LAW-19 Affirmation ≠ delivery

LAW-104 Session close protocol

LAW-112 Phase C pod order (binding)

SECURITY

LAW-50 No credential echo

LAW-101 No synthetic work

LAW-102 CVO registration gate

LAW-105 CI billing gate

LAW-107 AGT-WATCHDOG structural fix

LAW-111 Repository + filesystem hygiene

OPERATIONS

LAW-0§R Repo hygiene — constitutional

LAW-103 CVO source-of-truth

LAW-106 Session guard rails

LAW-108 Watchdog auto-cancel

LAW-109 R-records real lessons only

LAW-110 IAT tier honesty

API Reference

Base URL: https://ctx.orchestraios.com

All endpoints return JSON. No auth required for read endpoints in Phase C. Auth via IBCT tokens in Phase D.

GET

/context/orchestrai

Full OS state — all layers, agents, laws, metrics, and phase status in a single response.

// Response shape
{
  "layers": [...],
  "agents": [...],
  "laws": [...],
  "phase": "C",
  "version": "v1.1-ace-live",
  "metrics": { "layers_live": 25, "agents_operational": 21 }
}
            

GET

/context/layers

Layer list with ADLC flags, plane assignment, live status, and CVO registration metadata.

// Query params: ?plane=D&live=true&limit=20
{ "layers": [{ "id": "SL1", "name": "Sovereign LLM Gateway", "live": true, "plane": "S" }] }
            

GET

/context/agents

Agent list with IAT tier, operational status, heartbeat, and CVO registration info.

// Query params: ?iat_tier=1&status=operational
{ "agents": [{ "id": "AGT-ACELF01", "iat_tier": 1, "status": "operational" }] }
            

GET

/context/laws

All 75 constitutional laws with category, enforcement level, and description.

GET

/context/metrics

Phase metrics — layers_live, agents_operational, ip_claims, lessons count, eval coverage.

GET

/health

Service health check. Returns 200 OK when context API is up.

{ "status": "ok", "uptime": 99.7, "version": "v1.1" }

MCP Tools Reference

MCP endpoint: https://convo.orchestraios.com/mcp

All MCP tools require IBCT token authorization via mcp_auth_authorize_tool. Connect any MCP-compatible client (Claude Desktop, Claude Code, or custom agents).

cvo_query_readonly

Read-only SQL against the OrchestrAI CVO Postgres database. Returns layer, agent, law, and metrics data.

// Example
{ "sql": "SELECT id, name FROM cvo_agents WHERE iat_tier=1" }
        

Auth: IBCT token required · Rate: 100 req/min

mac_shell_execute

Execute shell commands on the OrchestrAI host. Used by agents for infrastructure operations and service management.

// Example
{ "command": "docker ps --filter name=orchestrai" }
        

Auth: IBCT token required · Tier: restricted

memory_search

Semantic search over OrchestrAI R-records (lessons). Returns relevant lessons from 192 real session-learned entries.

// Example
{ "query": "Constitutional Pod deployment failures", "limit": 5 }
        

Backed by: cvo_lessons (192 real R-records) · Qdrant

research_scan_tool

Web research via SearXNG (self-hosted). Runs searches across multiple engines and returns structured results.

// Example
{ "query": "DORA regulation MCP compliance 2024", "engines": ["google", "bing"] }
        

Backed by: SearXNG (self-hosted, privacy-preserving)

mcp_auth_authorize_tool

Issue an IBCT (Identity-Bound Capability Token) for authenticated MCP tool access. All other tools require a valid IBCT token.

// Step 1: Authorize
{ "client_id": "your-agent-id", "scope": ["cvo:read", "memory:read"] }
// Step 2: Use token in subsequent calls
{ "ibct_token": "eyJ...", "tool": "cvo_query_readonly", ... }
        

Token TTL: 1 hour · Scoped to declared capabilities

Evaluation Framework

OrchestrAI uses a 7-tier evaluation framework — from unit tests to domain-specific financial benchmarks. All tiers run in CI.

Tier	Name	Tool	Focus	Status
EV-1	Unit Tests	pytest	Agent function correctness, API contracts	● Live
EV-2	Integration Tests	pytest + docker	Agent-to-OS wiring, MCP tool calls, DB writes	● Live
EV-3	Constitutional Compliance	OPA + custom	75 law enforcement — every agent decision checked	● Live
EV-4	Formal Verification	Lean-4	Mathematical proof of compliance logic (33 theorems)	● Live
EV-5	Safety / Guardrail	Garak	Adversarial red-team, prompt injection, jailbreak resistance	● Live
EV-6	Quality Metrics	DeepEval	Faithfulness, relevance, coherence across agent outputs	● Live
EV-7	Domain Benchmark	IndiaFinBench	406 Q-A pairs across DORA, RBI, PCI-DSS, AML/KYC	● Live
EV-8	Prompt Regression	Promptfoo	Regression suite for all agent prompts — catches drift on model upgrades	● Live