One prompt, one job
vectr agent sends a natural-language prompt to the agent API, waits for the job, and prints either an answer or an unsigned transaction. Requires a logged-in key.
Updated
Usage
vectr agent "what did my launches earn this week"
vectr agent "build a 0.01 ETH buy of 0x…" --json
vectr agent "deeper analysis please" --model claude-opus-4-7| Flag | Effect |
|---|---|
| --json | Print the raw job result — status, response, richData — instead of formatted text. The scripting surface |
| --model <id> | Model override. The backend default is claude-sonnet-5, and an unknown id falls back to it |
- --json
- Effect
- Print the raw job result — status, response, richData — instead of formatted text. The scripting surface
- --model <id>
- Effect
- Model override. The backend default is claude-sonnet-5, and an unknown id falls back to it
What happens
The command posts an /agent/job, polls every 1.5 seconds until the job settles, then prints the response. When the job produced a transaction it prints a --- Transaction --- block with the action and details and a note that it is not signed. Submit it with a read-write key, or ask the same thing in the vectr.bot/terminal to review and confirm it there. The CLI never signs anything.
…agent's answer text…
--- Transaction ---
Action : Buy 0.01 ETH of EXAMPLE
Details: est. 12,400 EX at current curve price, 1% slippage
Not signed. Submit it from your wallet, or ask the same thing in the Vectr Terminal to confirm it there: vectr.bot/terminalThe example above is illustrative — token names, amounts and prices in it are not live market data.
Exit codes
| Code | Meaning |
|---|---|
| 0 | Job completed or was cancelled, or any finished job with --json (read status in the output) |
| 1 | Job failed (without --json), the request errored, or you're not logged in |
- 0
- Meaning
- Job completed or was cancelled, or any finished job with --json (read status in the output)
- 1
- Meaning
- Job failed (without --json), the request errored, or you're not logged in
Agent time is job time
A prompt is a job, not a request — it runs tool calls server-side until the turn settles. Prompts that trigger reads finish fast; ones that build transactions take as long as the agent needs to plan them.