Features · RunAIAgents

// features

A control plane for every stage of an agent's life

From the first conversational draft to a versioned production deployment you can watch in real time — RunAIAgents covers the whole loop, grouped by what you're trying to do.

// build

Design agents the way you'd describe them

Conversational builder: Describe the agent you want in plain language. RunAIAgents turns intent into a typed configuration you can refine — no YAML to memorize.
Visual canvas: See your agent as a live graph: the main agent, its sub-agents, tools, channels, and triggers, with animated data-flow between every node.
Marketplace: Fork a battle-tested agent template to start in minutes, or publish your own.

build · trace

01  prompt: "triage support tickets, refund < $50 auto"
02  → agent: support.triage
03  → tools: docs.search, mailer.reply
04  ✓ draft ready — open on canvas

// trigger

Five ways to start a run

Five trigger types: Run agents manually, on a schedule, from a signed webhook, on an external event (Slack/GitHub/Linear), or chained from another agent's output.
Human-in-the-loop: Gate sensitive tool calls behind an approval queue. Agents pause, a human approves or rejects, and the run resumes — fully audited.

trigger · trace

01  manual    → run.now()
02  schedule  → cron("0 7 * * 1")
03  webhook   → POST /hooks/:agent (signed)
04  event     → on("ticket.created")
05  chain     → after(agent.research)

// observe

Every run, recorded and replayable

Execution history: Every run is a complete, replayable trace — step by step, with inputs, outputs, token counts, and cost. Pin a step, re-run from any point, debug like n8n.
Versioning & rollback: Release immutable versions (V1, V2, …), diff what changed, and roll back instantly. Production never runs an un-frozen draft.
Privacy-first monitoring: An open-source sidecar reports only the metrics you consent to. Health scores, latency, cost, and anomaly alerts — without shipping your payloads anywhere.

observe · trace

01  run #4821  trigger=webhook  status=error
02    01 input.parse        ✓ 12ms
03    02 docs.search        ✓ 340ms
04  → 03 mailer.reply       ✗ 401 unauthorized
05    replay → step 03

// deploy

Run it where your data already lives

BYOC + BYOK deploy: Deploy to your own server over SSH (AWS, Oracle, Hetzner, DigitalOcean, Azure) with your own model keys. Your data and credentials never leave your infrastructure.
Managed hosting: Prefer us to run it? One click to managed hosting with metered, transparent billing and the same observability.

deploy · trace

01  target: byoc (ssh → your-cloud / us-east)
02  keys:   byok (provider=anthropic)
03  → provisioning runtime…
04  ✓ live — data residency: your account

See it on your own agent

Build an agent by conversation and run your first trace in minutes.

Start free View pricing