The agent calls read-only tools
It can review configuration, read metadata, sample objects, and import logs or inventory for DuckDB queries.
Storage Agent · For SRE & storage ops · Desktop · Read-only · S3-compatible
Local read-only storage investigations
Open a session about a bucket problem, error, or log. Storage Agent runs read-only checks on your machine and keeps the evidence beside the answer.
Desktop investigation agent
Chat plans the checks; evidence stays open
Describe a bucket, error, or log. The agent follows StorageOps skills, runs large analyses locally, and leaves the material it used beside the answer.
Example investigation
What it checked stays beside the answer
The agent compares object metadata and content, explains a storage-class mismatch, and leaves evidence you can open before the next step.
How investigations work
How investigations work
Follow a storage diagnosis skill
The agent loads a StorageOps skill that matches the problem and follows a known investigation path. Skills ship with the app; the agent does not fetch executable modules at runtime or run arbitrary shell commands.
Large analyses stay local
DuckDB and rule jobs process large datasets on your machine. The agent explains the result with supporting rows and tool records open beside it.
Answers list what was used — and what is missing
Each conclusion names skills used, evidence used, and gaps, so a fluent answer still points at material you can open.
How a session runs
How a session runs
Ask, or pick a starting task
Describe an error, bucket, log, or cost question — or start from config review, access logs, inventory, triage, or capacity.
The agent investigates with tools
It chooses read-only checks and keeps going under simple budgets for list, preview, and latency — so you are not confirming every small read.
Results land locally; you can open them
Analysis stays in local app data. Metadata, log rows, inventory, and tool activity stay next to the answer.
You approve larger actions
Downloads, large scans, and evidence imports pause for approval; ordinary metadata and preview reads stay within budgets. Then you decide the next step with the evidence in view.
Built to investigate safely
Built to investigate safely
Keys and evidence stay on your machine
Credentials live in an encrypted local vault; imported evidence, DuckDB state, and chat history stay in local app data. Requests go only to the model and S3-compatible storage endpoints you configure.
It cannot change your storage
There are no write, delete, or bucket-policy tools. The agent can investigate; it cannot mutate production.
Confirm downloads, large scans, and evidence imports
Ordinary reads stay fast. Downloads, large scans, and evidence imports always ask for approval.
What it is not
Not a full S3 browser or sync client, not multi-user SaaS, and not auto-fix or arbitrary shell.
Questions & how-to
How is this different from a console with chat?
Consoles manage objects; chat panels wrap APIs. Storage Agent runs the investigation: plan checks, call tools, analyze locally, and show evidence — without changing production storage.
Can it change my buckets or objects?
No. Access to configured storage is read-only. There are no write, delete, or config-change tools.
What if I have no model API key?
Local evidence and rule analysis can still run. Add a model when you want the agent to plan and explain the investigation in chat.
How is this different from StorageOps?
StorageOps adds storage skills to Pi. Storage Agent is the full desktop app: chat, tools, local analysis, evidence, and approvals in one place.