Define once.
Trust everywhere.
Repeatable AI SRE workflows that scale across your org—same config, same outcome, every time. AURA handles the 80% so your team can focus on the 20% that matters.
Stop rebuilding the same agent scaffolding every time
Platform teams are tasked with productionizing AI for other teams to consume. That means common standards, safe defaults, and integrations with existing model and tool infrastructure. The hard part isn't the idea—it's making it repeatable enough to scale.
AURA is the open-source agent harness built for this job. Define agents in a TOML manifest like a Kubernetes config. AURA ships the assembled solution —batteries included, production-hardened—so your teams aren't rebuilding scaffolding from scratch every quarter.
[llm]
model = "gpt-5.2"
[agent]
name = "Multi-Worker Orchestrator"
[orchestration.worker.sre]
description = "Incident analysis worker"
[mcp.servers.mezmo]
description = "Mezmo MCP server"
[mcp.servers.datadog]
description = "Datadog MCP server"
[mcp.servers.pagerduty]
description = "PagerDuty MCP server"
[[vector_stores]]
...From months to hours. From Lego blocks to an assembled solution.
Declarative workflows, production defaults
Define agents in a TOML manifest—like a Kubernetes config for AI. Open-source, works with any model or MCP server, config lives in source control. Same config, same outcome, anywhere.
Right data, right context, right cost
Sits between your telemetry sources and observability platforms. Filter, route, and enrich in-stream—no rip-and-replace. Teams see 50–85% cost reduction without changing what they're already running.
Control plane meets data plane
AURA is the car, Mezmo is the supercharger. Start with one, add the other when you're ready. Together, the same workflows become 50–70% more cost-efficient—agents get curated data, not a firehose.
One harness. Three workflow modes. Any model.
Agents run continuously, building contextual knowledge of your systems. Identify issues before they escalate and automate fixes nobody has time to address manually.
Automated triage, root cause analysis, and post-mortem generation. Same workflow every time—from alert to remediation in under 5 minutes.
Natural language access to production systems. Any team member can ask questions and get contextual answers—connected to real data, not a sandbox.
Mezmo's MCP tools are purpose-built for agent task completion—curated, right-sized telemetry for each query. No context window bloat. No hallucinations.
Everything your platform needs to run agents in production
Config in source control — testable, iterable, auditable. Same outcome anywhere.
Aha moment in under 1 hour. Full deployment in hours, not months.
Own your workflows and context. Works with any model, any MCP server.
Agents ask permission for sensitive operations. No accidental deletions.
Remove low-value data before it hits expensive destinations. Route signal.
Auto-scale processing capacity and sampling rates during traffic spikes.
Add contextual metadata automatically — agents work with richer inputs.
Redact, encrypt, or mask PII in-stream before it reaches any destination.
Explore more
Ship repeatable AI workflows in hours, not months.
- ✔ Schedule a 30-minute session
- ✔ No commitment required
- ✔ Free trial available
