What is AI-powered incident management?

AI-powered incident management uses intelligent agents to automate incident response tasks like alert triage, runbook execution, root cause analysis, and team notifications. Instead of relying solely on on-call engineers, AI agents can handle initial diagnosis and remediation steps 24/7, reducing MTTR and on-call burnout.

Is Akmatori open source?

Yes, Akmatori is an Apache 2.0 open-source project. You can inspect the code, run it on your own infrastructure using Docker or Kubernetes, and keep complete control over your data and incident workflows.

What LLM providers does Akmatori support?

Akmatori supports multiple LLM providers including OpenAI (GPT-4), Anthropic (Claude), Google (Gemini), and OpenRouter. You can also use on-premise models like Mistral, GLM, Kimi, or Minimax for data sovereignty requirements.

How does Akmatori integrate with PagerDuty?

Akmatori integrates with PagerDuty via webhooks and the PagerDuty API. When an incident is triggered, Akmatori's AI agents can automatically acknowledge alerts, gather context from your observability stack, execute diagnostic runbooks, and post updates to incident channels.

Can Akmatori reduce on-call burnout?

Yes. Akmatori handles routine incidents autonomously, filters alert noise, and only escalates to human engineers when necessary. This significantly reduces the number of pages during off-hours and allows SRE teams to focus on high-impact work instead of repetitive troubleshooting.

What observability tools does Akmatori work with?

Akmatori ingests alerts from Prometheus Alertmanager, Grafana, Datadog, Zabbix, and PagerDuty, and can monitor Slack channels. AI agents reach your infrastructure through real tools such as SSH and Kubernetes to diagnose issues.

Is Akmatori fully open source?

Yes. Akmatori is an Apache 2.0 open-source project. You can inspect the code, run it yourself, adapt it to your environment, and contribute improvements through GitHub.

How long does it take to set up?

Install takes a few commands: download the release docker-compose.yml and nginx config, then run docker compose up -d. Secrets are generated on the first run. Then add one alert source and one connector, such as SSH or Kubernetes, and send a test alert.

What should we connect first for a proof of concept?

Start with one noisy alert source, one production service, and the runbook or dashboard your responders already trust. Akmatori can then show a focused before-and-after: evidence collection, likely root cause, recommended remediation, and the approval step before anything changes.

Do we need to talk to sales before evaluating Akmatori?

No. Start with the Apache 2.0 codebase, run a self-hosted proof of concept, and validate Akmatori against your own alerts before planning a wider rollout. Enterprise conversations can come later when you are ready to review deployment architecture, integrations, governance, and support expectations.

Will the AI make changes to my production systems?

Only if you allow it. Akmatori supports multiple autonomy levels: from read-only diagnostics to fully automated remediation. You control exactly what actions the agent can take, with approval workflows for sensitive operations.

What LLM providers do you support?

Akmatori works with OpenAI, Anthropic, Google, OpenRouter, NVIDIA NIM, MiniMax, Ant Ling, and any custom OpenAI-compatible endpoint. You can also run fully on-premise with local models (GLM, Kimi, Mistral, LLaMA, vLLM, and similar).

How does Akmatori handle sensitive data?

Self-hosted deployments keep all data within your infrastructure. No telemetry, no external calls unless you configure an external LLM. For maximum security, run with local models and your data never leaves your network.

Can I integrate with my existing tools?

Yes. Built-in connectors cover Prometheus Alertmanager, Grafana, Zabbix, VictoriaMetrics, Datadog, Catchpoint, PagerDuty, Jira, Kubernetes, SSH, NetBox, PostgreSQL, ClickHouse, Slack, Telegram, and Discord. You can also connect any remote MCP server or HTTP API as a custom connector.

Akmatori | Open-Source AI Incident Response Platform for SRE Teams

Self-Hosted AI Incident Response for SRE, Platform, and DevOps Teams

Akmatori takes an alert from your monitoring, investigates it on your real servers, Kubernetes clusters, and dashboards, and tells you the root cause and the fix. It runs in your own infrastructure with the LLM you choose, and production changes wait for your approval.

Open-source Apache 2.0Deploy with Docker Compose or HelmWorks with your existing alerts and runbooks

Apache 2.0 and self-hosted. Clone it, run it on your own alerts, and keep every incident, runbook, and model inside your infrastructure.

See it work on a real incident

These screens come from our SRE Bench lab, where we break real servers and Kubernetes workloads on purpose and let the agent find out why.

A root cause you can act on

nginx returns 502 Bad Gateway. The agent finds the real cause one layer down: the billing-api config says PORT=31ab, so the service crash-loops. The summary gives the evidence, the exact fix, the checks to confirm it, and a warning not to restart nginx.

Akmatori incident summary: root cause of an HTTP 502 alert, with evidence and the fix

Every step in the open

The timeline shows each search, each SSH command, and the reasoning between them. The agent first checks runbooks and memory, then follows the evidence from the nginx error log to the failed service and its config file.

Akmatori incident timeline: SSH commands and the agent's reasoning during a live investigation

The same flow on Kubernetes

A pod cannot pull its image. The agent reads the pod detail, its events, and the deployment spec, and finds that the deployment points to an image tag that does not exist. The Kubernetes connector is read-only.

Akmatori incident view for a Kubernetes pod that cannot pull its image

Connect your stack once

Monitoring, Kubernetes, SSH, databases, ticketing, and chat, or any remote MCP server. Every enabled connector is available to every skill. Tools run in a separate gateway, so the agent never sees your credentials.

Akmatori connectors page: Kubernetes, SSH, NetBox, PostgreSQL, ClickHouse, Slack, Telegram, Discord, and custom MCP servers

Get Started in 60 Seconds

Deploy Akmatori locally, install it on Kubernetes, or trigger your first AI investigation from the CLI. No vendor lock-in, full data control.

# Clone and start Akmatori in under a minute
git clone https://github.com/akmatori/akmatori.git
cd akmatori
docker compose up -d

# Open http://localhost:8080 to access the UI

Quick start guide

Best for local evaluation and fast team demos

Boot the full stack on localhost with Docker Compose.
Open the UI immediately and connect your first alert source.
Validate incident workflows before touching production.

Prerequisites

Docker EngineDocker ComposeGit access

You are ready when

The web UI loads on localhost.
Core services show healthy in Docker Compose.
You can create a test incident from the dashboard.

Verify the stack is healthy

docker compose ps

You should see the Akmatori services in a running or healthy state before opening the UI.

Read Docker quick start View source on GitHub

# Trigger an AI-powered incident investigation curl -X POST https://akmatori.example.com/api/incidents \ -H "Authorization: Bearer $AKMATORI_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "task": "Investigate sustained CPU saturation on prod-web-01 and summarize likely root cause", "context": { "host": "prod-web-01", "severity": "critical", "source": "prometheus", "metric": "cpu_usage", "value": 98.5 } }' # Response: Akmatori starts investigating immediately # { # "id": "inc_abc123", # "status": "investigating", # "agent": "sre-agent-01" # }

AI incident response your SREs can operate

Akmatori turns a production alert into a finished first investigation: it reads your runbooks and memory, checks recent changes, runs diagnostic commands on your systems, and writes the root cause and the fix. Risky actions become proposals that an engineer approves. Self-host it, connect the tools you already use, and test it on your own alerts.

What Akmatori Does

The capabilities that ship in the open-source platform today

Multi-Source Alert Ingestion

Ingest webhooks from Alertmanager, PagerDuty, Grafana, Datadog, and Zabbix, plus Slack channel monitoring. Keep the monitoring you already have.

Agentic Investigation

Every incident runs a full agent session. It searches your runbooks and past incidents first, then collects logs, metrics, and state from your systems until it finds the root cause.

Connectors, Not Just Chat

Connect SSH, Kubernetes, monitoring, databases, ticketing, and remote MCP servers once. Tools run in a separate gateway, so the agent never sees your credentials.

Approval-Gated Remediation

Risky actions become proposals you approve in the web UI or Slack before anything executes. You decide how much autonomy each agent gets.

Bring Any LLM

OpenAI, Anthropic, Google, OpenRouter, NVIDIA NIM, MiniMax, or any OpenAI-compatible on-prem endpoint (GLM, Kimi, Mistral, LLaMA). Swap providers anytime — no lock-in.

Runbooks and Memory

Your team writes runbooks. After each incident, the agent saves what it learned to memory, which you can review, edit, or delete. Skills use the open Agent Skills format.

Self-Hosted by Design

Deploy with Docker Compose or Helm. Your incident data never leaves your infrastructure — run fully on-prem with local models for air-gapped setups.

Web Dashboard + Slack

Follow every investigation step by step in the web UI, run scheduled investigations with cron jobs, and get results in Slack, Telegram, or Discord.