Get running in 4 steps
1
Install the CLI
Get the AgentArc command-line tool.
npm i -g @agentarc/cli
2
Authenticate
Log in with your AgentArc account.
agentarc login
3
Create your first agent
Scaffold a new agent from a template.
agentarc new my-agent
4
Run it
Trigger a local run and see the trace.
agentarc run --watch

Core concepts

Understand agents, workflows, tools, and how they compose into production systems.

Read more →

Tools & memory

Connect APIs, databases, and external services. Give your agents persistent memory.

Read more →

Observability

Trace every step, token, and tool call. Replay failures and set up live alerts.

Read more →

Guardrails

Add policy checks, approval gates, and spend limits to keep every run safe.

Read more →

Scaling & queues

Auto-scaling workers, priority queues, and rate-aware routing for high-volume loads.

Read more →

Security & compliance

SSO, RBAC, audit logs, data residency, and SOC 2 Type II documentation.

Read more →
Jun 18
New
Parallel step execution
Workflows can now run independent steps in parallel with automatic fan-out and join. Up to 32 concurrent branches per workflow.
Jun 12
Improvement
Trace viewer — diff mode
Compare any two runs side-by-side in the trace viewer. Useful for debugging regressions between agent versions.
Jun 5
Fix
Memory retrieval on long contexts
Fixed a bug where vector memory retrieval returned stale results after 50k+ documents were indexed in a single namespace.
May 28
New
Anthropic Claude 4 support
All Claude 4 model variants are now available as agent backends. Set model: claude-opus-4 in your agent config.