Prerequisites
Before you begin, you'll need a Metaprise account and an API key. Sign up at metaprise.ai/signup — the free tier includes $1,000 in compute credits, no credit card required.
System requirements: Python 3.9+ or Node.js 18+. Any operating system. Internet connection for API calls.
Step 1 — Install the SDK
AURA SDKs are available for Python and TypeScript. Choose your language:
# Python
pip install metaprise
# TypeScript / Node.js
npm install @metaprise/sdk
# Verify installation
python -c "import metaprise; print(metaprise.__version__)"
Step 2 — Authenticate
Initialize the client with your API key. Keys are prefixed mp- for easy identification. Never commit API keys to version control — use environment variables.
from metaprise import AgentOS
import os
client = AgentOS(api_key=os.environ["MP_API_KEY"])
# Verify connection
print(client.ping()) # → "ok"
print(client.credits()) # → { "remaining": 1000.00, "currency": "USD" }
Step 3 — Define your first Mission
A Mission is AURA's fundamental execution primitive. You define what needs to be done — not how. The Runtime handles planning, model selection, tool execution, and audit.
A MissionDefinition has exactly four fields:
| Field | Type | Description |
|---|
objective | str | Human-language description of the goal |
tools | list[str] | Whitelist of permitted tool identifiers |
authority | str | Organizational unit that owns execution authority |
timeout | int | Maximum wall-clock seconds before escalation |
from metaprise import MissionDefinition
# Define a Mission — only 4 fields, no steps, no model names
mission = MissionDefinition(
objective = "Reconcile outstanding invoices for Q1 2026",
tools = ["accounting_api", "contract_db", "email"],
authority = "finance_team",
timeout = 3600
)
# Launch the Mission
result = client.missions.launch(
definition = mission,
context = {"quarter": "Q1-2026", "entity": "ACME-US"}
)
Key insight: You don't write workflow steps. The Orchestrator decomposes your objective into a plan, the Runtime executes it, and the AuditChain records every action. You write what — AURA handles how.
Step 4 — Observe the Mission lifecycle
Every Mission passes through 8 states. No state can be skipped. Every transition writes an AuditChain record.
CREATED → PLANNING → WAITING_APPROVAL → APPROVED
→ PENDING_EXECUTION → IN_PROGRESS → EXECUTED → CLOSED
# Watch the Mission execute
for event in client.missions.stream(result.mission_id):
print(f"[{event.state}] {event.message}")
# Output:
# [CREATED] Mission msn_7f3k9... created
# [PLANNING] Decomposing objective into 3 steps
# [APPROVED] Auto-approved (L1 authority)
# [IN_PROGRESS] Step 1/3: read_invoice
# [IN_PROGRESS] Step 2/3: verify_contract
# [IN_PROGRESS] Step 3/3: create_payment_draft
# [EXECUTED] All steps completed
# [CLOSED] Mission outcome finalized
Step 5 — Verify the audit trail
Every Mission produces a tamper-proof AuditChain. Each entry is cryptographically linked to the previous, making retroactive modification detectable.
# Get the completed outcome
outcome = client.missions.get_outcome(result.mission_id)
print(outcome.status) # → "SUCCESS"
print(outcome.actions_taken) # → ["read_invoice:4521", ...]
print(outcome.cost_usd) # → 0.12
print(outcome.duration_ms) # → 18430
# Verify audit chain integrity
chain = client.audit.get_chain(outcome.audit_chain_id)
assert chain.verify_integrity() # → True
# Every step is independently verifiable
for entry in chain.entries:
print(f" {entry.event}: {entry.tool} → {entry.result}")
Step 6 — Publish to the Mission Registry
Once your Mission works, publish it so enterprises can discover and install it. You'll earn 30% of every execution fee.
from metaprise import MissionPackage,
MissionRegistry
package =
MissionPackage(
name =
"invoice_reconciliation",
version =
"1.0.0",
definition = mission,
description =
"Reconcile invoices against contracts",
tags = [
"finance",
"reconciliation"],
royalty_pct = 25
)
registry =
MissionRegistry(org_id=
"acme-corp")
registry.
publish(package)
# → Published [email protected]
Next steps
You've launched your first Mission, verified its audit trail, and published it to the Registry. Here's where to go next:
- Python SDK Reference — Complete API documentation for the Python SDK
- Mission Registry — Learn about versioning, forking, and the royalty chain
- OrgKernel — Integrate enterprise trust into your existing agents
- Examples & Tutorials — Compliance review, multi-agent workflows, and more
- Developer Economy — Understand the revenue share model and how to maximize earnings