opendevops

Documentation

Build, operate, and reason about an autonomous operations agent whose authority stays narrower than its prompt.

opendevops is an open-source operations agent built on LangGraph and LangChain deepagents. It can investigate infrastructure, diagnose incidents, explain CI failures, verify drift, and execute a deliberately narrow set of remediations.

The model never receives a shell. Every command is a structured argv request that crosses budget controls, fail-closed policy, credential selection, output scrubbing, and a per-run audit chain.

Choose your path

What is supported

AreaSupported todayBoundary
Kubernetesdiagnostics, logs, events, Helm inspection, controlled apply/rollout/scaleproduction write requires an active expiring grant, server dry-run, independent approval, and a write credential
GitHubCI diagnosis, run inspection, pull-request remediationrepository and API method/path allowlists
AWScurated EC2, ECS, RDS, CloudFormation, S3, Lambda, CloudWatch, and related readsno cloud-resource deployment or IAM access
Google Cloudcurated Compute, GKE, Cloud SQL, Pub/Sub, Logging, Storage, Run, and Functions readsmutations and secret access denied
Azurecurated VM, AKS, ACR, networking, SQL, Cosmos DB, Monitor, and resource readsmutations and secret material denied
Remote hostsstructured, read-only SSH checkspinned user, key, hosts, and known_hosts
InterfacesCLI, HTTP, Slack, scheduler, Alertmanager, GitHub webhooks, operations UIevery path enters one gateway and safety core

opendevops is not a general cloud deployment engine. Terraform, Pulumi, provider-wide deployment commands, IAM mutation, and unrestricted CLI access are intentionally unavailable.

The operating invariant

The model may propose an action. It cannot make the action permissible.

Permission emerges only when all independent controls agree:

  1. the active policy pack recognizes the exact tool, executable, subcommand, flags, target, and environment;
  2. the request is still within its run and daily budgets;
  3. a scoped credential is configured for the winning policy rule and execution channel;
  4. dangerous actions have a typed, unexpired capability grant with execution capacity remaining;
  5. required dry-run and human approval checks have completed;
  6. the executor accepts the signed, structured request.

Continue with Getting started, or read the architecture before you connect infrastructure.

On this page