AI that does the boring part.
No chat sidebar you have to babysit. Five features, each invoked by keystroke, each BYOK.
Natural language to $SHELL.
Type "find all files bigger than 1GB modified in the last week" — vev drops the find incantation into your prompt, highlighted, waiting for you to hit return. You read it first. It runs second. Works for shell, git, kubectl, Terraform HCL, jq, SQL. All of it.
Paste a stack trace. Get a hypothesis.
Select a log region, hit ⌘L. vev summarizes the failure mode, names the likely root cause, points you at the log line that matters. No chat thread. One answer, one scroll.
Describe infra. Get Terraform.
"A Postgres 15 on RDS, multi-AZ, in eu-west-1, with a dev read-replica." vev writes the .tf, names the resources sensibly, flags the variables you need to fill. You review. You plan. You apply. Nothing leaves your machine that you didn't send.
One buffer. Shell, SQL, Python, AI cells.
Mix them. Chain them. The SQL output of cell 2 is a dataframe in cell 3. The AI in cell 4 reads both. Export as markdown. Commit to the repo.
You bring the key. We bring the wire.
Anthropic, OpenAI, Groq, Gemini, any OpenAI-compatible endpoint (Ollama, LM Studio, vLLM). Your key is stored in your OS keychain. It never touches our servers because we don't have servers for this.
the contract
- — Your prompts go directly to the provider you configured.
- — We log nothing. We route nothing.
- — Pull mitmproxy on our binary. You'll see.