From copilot to autonomous incident response.
Three phases, each building on real feedback from beta teams. No dates until we're sure — just honest status labels.
Phase 1 — Private Beta
Novix Flow as an MCP server inside your Claude client. Real incidents, real context, human in the loop.
ClickHouse error ingestion
query_clickhouse_errors finds error spikes by service and time window, straight from your telemetry tables.
Linear context
get_linear_context pulls related issues for the affected service so Claude reasons with your team's history.
Repository context
read_repo_files gives Claude the exact source files involved in the failure.
Incident triage workflow
The triage-incident MCP prompt guides Claude from spike → root-cause hypothesis → proposed patch.
Linear write-back
update_linear_issue creates or updates the issue with the root cause and the proposed patch.
Local-first
Runs on your machine with your own Claude client. Novix stores no code or telemetry.
Phase 2 — Autonomous Agent
The agent runs on its own: Haiku watches every spike, Sonnet handles the hard ones, and no patch reaches a human without passing your tests.
Autonomous agent mode
Runs without a human in the chat: Claude Haiku triages every spike and escalates complex multi-file cases to Claude Sonnet.
Prompt-cached repository context
Keeps your repo's structure and hot files warm in Claude's prompt cache for fast, cheap repeated analysis.
Verified patches
Every proposed patch will run your test suite in an isolated sandbox before a PR is opened.
GitHub PR automation
Opens a PR with root-cause summary, linked Linear issue and test results attached.
Sentry connector
A second telemetry source alongside ClickHouse.
Phase 3 — Novix Cloud & Enterprise
A hosted incident command center: multi-agent investigations, deploy correlation, postmortems written for you, and the controls enterprises need.
Novix Cloud
Hosted platform: no local setup, shared team workspace, always-on monitoring.
Incident Command Dashboard
Live incident map, MTTR trends, model routing and token analytics in one view.
Ephemeral microVM verification
Firecracker-based sandboxes running full E2E (Playwright) suites for every patch.
Multi-agent incident response
Parallel investigator agents for logs, traces, recent deploys and dependency changes, coordinated by an orchestrator.
Deploy correlation
Links every spike to the exact commit or deploy that caused it, with a one-click rollback proposal.
Postmortem generator
Automatic timeline, root cause, impact and action items, written to Linear or Notion.
More connectors
Datadog, Grafana / Prometheus, PagerDuty, Slack, GitLab and Jira.
Policy engine
Per-repo rules: what the agent may touch, approval levels, budget caps.
Enterprise
SSO, audit logs, VPC deployment and a SOC 2 program.
Want to shape what we build next?
Join the private beta