About Aura
AURA is an open source harness from Mezmo for building production AI agents that handle operations work alongside engineering teams. Agents are composed from declarative TOML configuration: a team picks an LLM provider, attaches vector stores such as runbook collections in Qdrant, connects tools through MCP servers, and defines an orchestrator that routes work to specialist agents. The framework is written in Rust and exposes an OpenAI compatible streaming API.
It is aimed at platform and site reliability teams. A typical configuration has an SRE orchestrator decompose an incident and delegate in parallel to an incident responder using PagerDuty, a metrics analyst querying Prometheus and a log analyst searching and correlating log events, with each worker limited to a filtered set of tools and a bounded number of turns. AURA is deliberately agnostic: it works with any model provider, any MCP tool and any observability backend, and telemetry goes wherever the team routes it through OpenTelemetry. Mezmo says it is the same framework that powers its own agentic SRE features.
AURA is free under the Apache 2.0 license on GitHub, with community help through Slack. Mezmo's commercial telemetry pipeline and AI SRE platform, which pairs naturally with AURA, does not publish prices; the pricing page explains that AI root cause analysis is included in the platform licence without per query charges and directs buyers to talk to an engineer for a quote.
Mezmo describes AURA as open core, with the company as primary contributor and steward of the project.