Qredence Documentation

The CLI is the fastest way to script the agent or run one-shot tasks. It shares the same runtime as the TUI and web app.

Modes

Pass a mode with the first positional argument or --mode:

text (default) — interactive text mode.
json — emit one JSON ChatStreamEvent per line on stdout. Good for pipelines.
rpc — JSON-RPC framing over stdio. For building your own client.
daemon — start the background daemon.
acp — Agent Communication Protocol mode.

Flag	Purpose
`--provider`	Provider id, e.g. `openai`, `anthropic`.
`--model`	Model id from the provider.
`--api-key`	Inline API key, kept in memory only (not persisted).
`--thinking`	Thinking level: `off`, `minimal`, `low`, `medium`, `high`, `xhigh`, `max`.
`--autonomous`	Run without pausing for interactive questions.
`--resume <id>`	Resume a session by id.
`--fork <id>`	Fork a session by id.
`--extension <pkg>`	Load an extension package.
`--skill <path>`	Load a skill package from a path.
`--theme <name>`	Set the terminal theme.
`--no-tools`	Disable all tools for this run.
`--no-extensions`	Disable all extensions for this run.
`--no-env`	Ignore ambient environment variables (`OPENAI_API_KEY`, etc.).
`--goal <text>`	Set an explicit goal for the run.
`--goal-token-budget <n>`	Cap tokens spent on the goal.

--list-models was removed. Use prime-agent model list instead.

One-shot summarization piped from stdin:

cat report.md | prime-agent --provider anthropic --model claude-sonnet-4 --autonomous "Summarize this report in five bullets"

Resume and pipe events to jq:

prime-agent --resume abc123 --mode json | jq -rc 'select(.type=="delta") | .text'

Run without any tools (pure chat):

prime-agent --no-tools --provider openai --model gpt-4o "..."

See the source of truth for flags: packages/coding-agent/src/cli/args.ts.

Last modified on August 17, 2026

⌘I