Arbiter is an MCP-native AI agent that triages IT tickets and decides to auto-resolve or escalate based on trust scores and hard-coded risk keywords (security, billing, and production) that force human review. Unlike Jira-Slack integrations that just move data, Arbiter makes decisions. Built with vendor-agnostic interfaces where Jira, Slack, and Groq are reference implementations, not dependencies. Apache 2.0, open for contributors.
How did Astra change the scope or ambition of what you built?
Maker
Astra allowed us to push our reasoning architecture further by enabling more reliable, low-latency agent context handling, which directly improved the precision of our automated trust-scoring and escalation loops.
Report
Maker
📌
Maker here 👋 Built this because most "AI ticket triage" tools I found just relabel tickets — none of them commit to a decision.
Arbiter will actually auto-close a ticket if (and only if) the trust score clears the threshold and no risk keyword fires.
Happy to answer anything about the scoring design or the benchmark methodology (40 hand-graded tickets, including deliberately tricky ones).