Use AI coding agents
where you normally can’t.
AutoDevOps captures what your AI agents do — tool calls, model usage, approval decisions — across your laptop, CI, and cloud workers. Your security team gets the audit trail their regulators require. You get to use agents without being watched. Pre-production only, inside your own cloud.
Every agent. On the record.
Connectors live and shipping. Coverage detail at /directory.
Agent actions only
No keystrokes, no screen, no editor
Pre-production only
Your laptop, CI, and cloud workers
Your cloud, your keys
Nothing leaves the perimeter
Allow · Confirm · Block
Routine work flows; risk pauses
On the record — without being in your way.
AutoDevOps observes the agent, not the developer. Three things happen between an agent deciding to act and that action being allowed:
Step 1
Capture
We sit in your hook chain at commit, push, and CI. Agent tool calls and model usage flow through. Your editor, browser, and keystrokes never do.
Step 2
Decide
Routine work passes silently — same as today. Destructive commands, external network calls, and policy-flagged actions pause for approval. Same gate on your laptop, in CI, and on cloud workers.
Step 3
Record
Every decision is signed, immutable, and stored inside your own cloud. Auditors get one system of record. You never have to reconstruct your day.
One config. Same gate everywhere.
The same policy file governs agents on your laptop, in CI, and on cloud workers. Start on one repo; roll it across the org when the controls earn trust.
hooks:
pre-commit:
- run: lint
- run: security-scan
block_on: critical
governance:
default: confirm
approvals:
require_for:
- destructive_commands
- external_network
providers:
bedrock:
region: us-east-1
budgets:
per_commit_tokens: 5000Policy engine
- Deterministic rules — no LLM interpreting your policy
- Once an action is approved, similar future actions can flow
- Optional: the agent self-assesses before a human is asked
Audit trail
- Every agent decision replayable with full context
- What the agent saw, decided, and did — and who approved it
- Same record whether the agent ran on a laptop or a cloud worker
Budgets, not surprises
Token and cost limits per commit, per repo, per team. Hit a limit and the next agent call pauses for approval — your finance team isn’t the one finding it on the invoice.
Inside your cloud. Owned by you.
There is no AutoDevOps SaaS plane that holds your prompts, source, or audit log. The whole control plane runs in your AWS, Azure, or GCP account, under your IAM, in your VPC.
AWSLive
Bedrock for inference, IAM-scoped Lambda workers, VPC endpoints, immutable audit storage. Prompts and source never leave your account.
Microsoft AzureSupported
Azure OpenAI, Functions in private networks, Key Vault, and the same audit contract. Your tenant, your keys, your data residency.
Google CloudSupported
Vertex AI for inference, Cloud Run workers, Secret Manager, and Firestore for state. Same policy engine, your perimeter.
Whatever your developers already use
The agent connector is the same shape regardless of vendor.
See what your auditor will see.
A live walkthrough on a real workflow — policy, approvals, audit — deployed into your own AWS, Azure, or GCP account so your security team can poke at it.