Documentation de l'API Agent d'Insight Profond

v1 · Authentification de base + envoi de transcriptions + lecture des analyses

Authentification

Toutes les interfaces /api/v1/* nécessitent une clé API dans l'en-tête de la requête :

Authorization: Bearer lj_live_xxxxxxxxxxxxxxxx

Les clés API se créent dans /settings/api-keys.

POST /api/v1/transcripts

Envoyer du texte transcrit (cas d'usage : appareils matériels, webhooks d'outils d'enregistrement tiers)

Requête

curl -X POST https://deepagent.wiki/api/v1/transcripts \
  -H "Authorization: Bearer lj_live_xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "说话人1: ...\n说话人2: ...",
    "title": "周三战略会议",
    "device_id": "meeting-room-shanghai-01"
  }'

Réponse

{
  "transcript_id": "550e8400-e29b-41d4-a716-446655440000",
  "analysis_id": null,
  "message": "转写已保存"
}

GET /api/v1/analyses

Lister les analyses récentes de l'org (50 éléments)

curl https://deepagent.wiki/api/v1/analyses \
  -H "Authorization: Bearer lj_live_xxxxxxxx"

GET /api/v1/analyses?id=xxx

Obtenir le détail d'une analyse (avec les ancres de preuves)

curl "https://deepagent.wiki/api/v1/analyses?id=ANALYSIS_ID" \
  -H "Authorization: Bearer lj_live_xxxxxxxx"

POST /api/v1/ask

Ask your second brain. The system auto-picks quick RAG or deep agent retrieval; streams via SSE.

curl -N -X POST https://deepagent.wiki/api/v1/ask \
  -H "Authorization: Bearer lj_live_xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "上次客户的核心顾虑是什么？我们答应了什么？",
    "projectId": "可选：限定某个项目",
    "mode": "auto"
  }'

SSE events: mode / tool_call·tool_result (agent steps) / sources / token (streamed answer) / insufficient / done / error.

data: {"type":"mode","mode":"agent"}
data: {"type":"tool_call","tool":"search_atoms","args":{"query":"客户 顾虑"}}
data: {"type":"tool_result","tool":"search_atoms","result":"[#1] (人物判断) …"}
data: {"type":"token","delta":"客户的核心顾虑集中在…"}
data: {"type":"done"}

Connect your brain (MCP)

Expose your second brain as an MCP server — any MCP-capable agent (Claude Desktop, Cursor, …) can call your org knowledge as tools.

MCP endpoint：https://deepagent.wiki/api/mcp（Authenticated with your API key · requires brain.ask scope）

Exposes these tools:

• ask_brain(question, project_id?) — Ask the brain — full agent, multi-step retrieval with citations
• search_brain(query, project_id?) — Semantic search over sedimented insight atoms
• list_decisions(project_id?) — List decisions and action items with status
• ingest_transcript(text, title?) — Reverse ingestion — feed a transcript straight into the brain (requires transcript.write)

Typical flow: a DingTalk/Feishu bot receives a meeting transcript and calls ingest_transcript to push it into your second brain — analyzable and askable from then on.

Claude Code

claude mcp add --transport http deepbrain https://deepagent.wiki/api/mcp \
  --header "Authorization: Bearer lj_live_xxxxxxxx"

Claude Desktop（claude_desktop_config.json）

{
  "mcpServers": {
    "deepbrain": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://deepagent.wiki/api/mcp",
               "--header", "Authorization: Bearer lj_live_xxxxxxxx"]
    }
  }
}

Cursor（.cursor/mcp.json）

{
  "mcpServers": {
    "deepbrain": {
      "url": "https://deepagent.wiki/api/mcp",
      "headers": { "Authorization": "Bearer lj_live_xxxxxxxx" }
    }
  }
}

Restart your client after setup, then ask it to pull what your brain knows about a topic.

Limitation de débit

60 requêtes par minute / clé API (au-delà, renvoie 429)

Permissions (scopes)

• transcript.write — Envoyer des transcriptions
• analysis.read — Lire les analyses
• brain.ask — brain.ask — ask & search the brain (for /api/v1/ask and MCP)
• * — Toutes

Codes d'erreur

Support

Signalement de problèmes :hi@zaowuyun.com