This document describes promethea CLI as a protocol client of the runtime.
CLI is both:
- A human-facing front end (terminal-first assistant usage).
- An automation/protocol harness (for scripts, CI, and agent-to-agent calls).
CLI is not a separate runtime. The backend runtime remains the source of truth.
- You want to use Promethea without the Web UI.
- You want stable commands for smoke tests and CI.
- You want to confirm that protocol surfaces are actually callable.
# auth
promethea auth register <username> <password>
promethea auth login <username> <password>
# one chat turn
promethea chat send "请���天�工�" --stream
# inspect tools and services
promethea status official-tools
promethea status serviceschat sendchat chat-confirmaskfollowupsessions listsessions show <session_id>
Use this group when your goal is "talk to the agent now".
config getconfig effectiveconfig updateconfig resetconfig switch-modelconfig ui-schemaconfig contractconfig runtimeconfig runtime-scoped
Use this group when your goal is "change behavior".
memory capabilitiesmemory graphmemory entries-listmemory entries-creatememory entries-updatememory entries-deletememory recall-runsmemory recall-inspect
Use this group when your goal is "inspect or maintain long-term memory".
workflow defineworkflow listworkflow startworkflow runworkflow pauseworkflow resumeworkflow retryworkflow approveworkflow checkpoints
Use this group when your goal is "long-running or resumable tasks".
call(generic tool invoke)status toolsstatus official-toolsops capabilitiesops protocolops methodsops http-contractsops surfacesops runbook
Use this group when your goal is "what can this runtime do right now".
Experimental/provider-dependent in the current preview. These commands are not part of the supported DeepSeek-only setup path because DeepSeek chat APIs do not provide STT/audio transcription.
voice capabilitiesvoice stt <file>voice tts <text>voice turn <text>voice ptt <file>
Use this group only when an OpenAI-compatible audio transcription provider is configured.
promethea chat send "请è°�ç � OpenAI æ��æ�°æ¨¡å��ï¼�å¹¶å�¨ workspace å��ä¸�份 markdown æ�¥å��" --stream
promethea workflow list
promethea status toolspromethea ops surfaces
promethea ops protocol
promethea ops http-contracts
promethea doctor runpromethea memory entries-list --limit 20
promethea memory recall-runs --limit 10
promethea memory recall-inspect <request_id>CLI can drive almost all backend capabilities already exposed by HTTP/WS contracts.
Still UI-heavy today:
- Dense visual graph exploration.
- Highly interactive drag/drop or modal workflows.
- Theme/avatar UX presentation.
These are presentation differences, not core capability gaps.
- API keys and provider secrets remain env-managed.
- Side-effect tools may require approval depending on policy.
- Dependency state (MCP/Neo4j/provider reachability) changes callable surface.
- CLI output is useful for both humans and machine parsing, but for strict contracts prefer
ops http-contractsand route schemas.
- If command exists but fails: run
promethea status servicesfirst. - If tool call denied: inspect policy using
promethea config tool-policy. - If memory behavior odd: inspect
memory recall-runs+memory recall-inspect. - If voice fails: confirm that an audio transcription provider is configured; DeepSeek-only chat configuration is not enough for voice input.
For CI or release gate, keep these as minimal checks:
- one chat turn
- official tools list
- workflow start + run
- config effective
- ops protocol + http-contracts