Step 1 of 7

Define Your Agent's Identity.

Give your agent a name, a clear purpose, and a visual identity. This shapes how it presents itself to your team and how it describes its own capabilities.

Basic Identity
Keep this tight — 1-2 sentences max. 0/200
Agent Icon
🤖
🔮
🧠
⚡
🔍
📊
🛡️
🗿
⚙️
📡
🔗
🏗️
Audience
Step 1 / 7
Step 2 of 7

Configure the Brain.

Choose the model, define the system prompt, and tune persona and tone. This determines how your agent reasons, communicates, and handles uncertainty.

Model Selection
Recommended
Claude Sonnet 4.6
Best balance of speed, reasoning, and cost. Ideal for most business agents.
Highest Capability
Claude Opus 4.6
Maximum reasoning depth. Best for complex analysis and architecture decisions.
Fastest / Cheapest
Claude Haiku 4.5
High-volume tasks, triage, routing, and real-time responses.
OpenAI
GPT-4o
Alternative when existing OpenAI keys or workflows are in place.
AWS Native
Bedrock Nova Pro
Stays entirely within your AWS account. No external API calls.
Custom
Fine-Tuned Model
Bring your own fine-tuned or self-hosted model endpoint.
System Prompt
Variables: {company_name}, {user_name}, {current_date}, {obelisk_context} 0 chars
Persona & Behavior
Precise Creative 0.3
Lower = more deterministic. Higher = more exploratory. Most agents: 0.2–0.5.
Step 2 / 7
Step 3 of 7

Connect the Obelisk.

Define what your agent knows. Connect it to your Data Obelisk layers and any additional knowledge sources. Grounded agents beat hallucinating ones every time.

Obelisk Layers
🗿
Past Layer — Institutional Memory
Decisions, incidents, post-mortems, vendor outcomes, technical debt log
📡
Present Layer — Live Context
Current OKRs, cloud cost baseline, CI/CD state, team structure, compliance posture
🔭
Future Layer — Strategic Trajectory
Roadmaps, hiring plans, product milestones, market positioning
Additional Sources
📄
Document Library
Runbooks, SOPs, architecture docs, compliance policies — uploaded PDFs / Markdown
🗃️
Confluence / Notion
Synced wiki pages via OAuth — updated automatically on a schedule
💰
AWS Cost & Usage Reports
Direct S3 CUR integration — daily cost data indexed into the agent context
🔐
Security & Compliance Findings
AWS Security Hub, GuardDuty alerts, SOC 2 audit trail
RAG Configuration
Strict mode recommended for compliance, cost, and security agents.
Step 3 / 7
Step 4 of 7

Wire the Tools.

Connect your agent to external systems via MCP. These give it the ability to act — not just answer. Select only what this agent actually needs.

Communication
✓
💬
Slack
Post messages, create channels, read threads, tag users
✓
📧
Email / SES
Send formatted email reports and alerts via AWS SES
✓
🔔
SNS / PagerDuty
Trigger alerts, escalations, and on-call paging
Project & Ticketing
✓
🎯
Jira
Create tickets, update status, assign epics, query sprints
✓
🐙
GitHub
Open issues, review PRs, query repos, read CI/CD status
✓
📋
Linear
Issue tracking, sprint management, roadmap updates
Cloud & Infrastructure
✓
☁️
AWS Cost Explorer
Query spend, forecasts, reserved instances, savings plans
✓
📊
CloudWatch
Read metrics, alarms, log insights, dashboards
✓
🛡️
Security Hub
Query findings, compliance status, GuardDuty alerts
✓
🏗️
Terraform Cloud
Trigger plans, query state, review workspace runs
✓
🔧
AWS Systems Manager
Run commands, query Parameter Store, manage patches
✓
📈
Datadog / Grafana
Query APM, traces, dashboards, SLO status
Business Systems
✓
💼
Salesforce
Query accounts, opportunities, pipeline, customer data
✓
🌐
Web Search
Real-time web lookups for pricing, docs, CVEs
✓
🔌
Custom API
Define a custom REST or GraphQL endpoint via MCP spec
Step 4 / 7
Step 5 of 7

Build the Cowork Network.

Define which other agents collaborate with this one, how handoffs work, and what triggers cross-agent workflows. Multi-agent systems multiply impact without multiplying headcount.

◆ Agent Flow Preview
🤖
Your Agent
Collaborating Agents
📊
Cost Sentinel
Cloud Cost Analyst · Always-on
🛡️
Compliance Warden
Security & Compliance · On trigger
Orchestration Mode
Conditional and Supervisor modes require an intent classifier step to be configured.
Step 5 / 7
Step 6 of 7

Set the Cadence.

Define when and how your agent activates. Agents can be on-demand, event-driven, or scheduled — or all three. Background workers run without anyone asking.

Activation Mode
✓
💬
On Demand
User queries the agent directly via chat, Slack, or API
✓
🕐
Scheduled
Runs automatically on a cron schedule in the background
✓
⚡
Event Driven
Triggers on webhook, CloudWatch alarm, or SNS message
Scheduled Runs
This prompt runs autonomously on schedule. Be specific about what to check and where to send output.
Event Triggers
💰
Budget Threshold Alert
AWS Budgets SNS → agent activates when spend crosses defined %
🚨
CloudWatch Alarm
Agent activates when a metric alarm transitions to ALARM state
🔗
Webhook / API Call
External system POSTs to the agent's endpoint to trigger a task
🐙
GitHub PR / Push Event
Agent reviews code changes, checks IaC compliance, or runs security scan
Step 6 / 7
Step 7 of 7

Review & Deploy.

Confirm your agent's configuration before launching. Everything can be edited after deployment — but get the core right before it goes live.

Identity
—
—
Brain
Claude Sonnet 4.6
—
Knowledge
Obelisk Connected
Past + Present layers active
Tools
No tools selected
MCP integrations
Cowork Agents
2 Collaborators
Cost Sentinel · Compliance Warden
Activation
On Demand
Chat, Slack, API
Ready to Deploy

Your agent will be deployed to AWS Bedrock via Lambda. It will be accessible via API, Slack bot, and the Nycolai Sage interface immediately after launch.

Deployment connects to AWS Bedrock via Lambda. Export Skill File downloads a SKILL.md you can drop into any Claude Code project.

Step 7 / 7