Build, run, and govern AI agent systems
The featured guide is the end-to-end walkthrough for building a multi-agent workflow system on Nagent. Platform documentation and developer reference for every other surface is being published in parallel.
Start here
End-to-end how-to guides. The multi-agent workflow guide is the first published; running individual agents and best practices are next on the queue.
Complete guide: building a multi-agent workflow system
The end-to-end how-to. From a one-sentence Helix brief to a governed, observable, self-improving multi-agent system on the substrate.
- 8-step build path
- Helix synthesis to deployment
- Governance + budget caps
- Sandbox evals + Karmic feedback
Running individual agents
Deploy a single agent — configure guardrails, hook triggers, observe runs, and iterate. For when you don't need a full multi-agent workflow.
- Agent lifecycle
- Triggers + guardrails
- Observability tab walkthrough
- Iteration patterns
Best practices for agentic systems
Field-tested patterns for autonomy tuning, oversight design, prompt versioning, eval suites, and cost governance from production deployments.
- Autonomy ladder tuning
- Per-action override design
- Eval-driven skill versioning
- Cost governance patterns
Every layer of the Nagent platform
Reference documentation for the Agent Control Plane, Workbench, Topology, Smriti, Karmic, Helix, BuildCraft, and the governance substrate. While these are being written, theTechnology Overview covers every layer in narrative form.
Agent Control Plane
The live operator surface — SSE stream of every action, decisions ledger replay, kill switch, materialised view at fleet scale.
Agent Workbench
Per-agent tuning surface — Tools / Triggers / Guardrails / Skills / Budget / Sandbox / Observability tabs and rollback.
Topology
Living-systems visualisation. Cluster agents into SYSTEMs, see cognitive gravity, matrix view, time-everywhere replay.
Smriti — memory fabric
Persistent per-entity memory. Cross-agent + cross-thread continuity, structured sections, continuous synthesis.
Karmic — outcome feedback
Outcome-linked attribution, per-agent trust scores, auto-demote on drift, template + prompt promotion.
Governance + autonomy ladder
L0–L4 autonomy, per-action overrides, kill switch, multi-party approvals, budget caps, JIT access, MFA.
Helix — Agent Studio
Chat-first agent + multi-agent-workflow synthesis. Builder UX, AgentSpec contract, auto-fixer, eval attachment.
BuildCraft — pro-code SDK
TypeScript + Python SDK with Git-native workflow. 9 modules (Spec · Tools · Skills · Triggers · Data/RAG · Governance · Observability · UI · Deployment).
API + SDK + integration surfaces
The traditional reference docs — request/response shapes, error codes, webhook signatures, auth flows. Generated from the live OpenAPI spec, publishing shortly.
Quickstart
10-minute walkthrough — API key to first webhook delivery.
Agent SDK
Python + TypeScript SDKs, type defs, worked examples.
REST API reference
Endpoint reference, request/response shapes, rate limits.
Webhooks
Real-time event subscriptions + signature verification.
Authentication
API keys, OAuth, SAML SSO, rate limits, rotation.
Concepts
How Nagent works under the hood — substrate primitives.
Need help building?
Our solutions team will pair-program with yours to get the first agent live in days, not weeks — and walk through the 8-step path on your actual workflow.
Talk to a solutions engineer