CrowTerminal - External Brain for AI Agents
"Agents are ephemeral. We are persistent."
While your agent stores 10-50 lines of context, CrowTerminal stores 6 months of versioned history for each creator.
What It Does
CrowTerminal is a persistent memory layer for AI agents working with influencers/creators:
- - Versioned Memory - Track what works across sessions (hook patterns, engagement, posting times)
- Pattern Detection - See trends over months, not single data points
- Engagement Analysis - Know what configuration performed best historically
- Validation - Check if your changes will repeat past mistakes
- Data Ingestion - Push platform data we can't access (retention curves, demographics)
- LLM-Native API - Schema discovery, semantic field aliases, natural language queries
Quick Start
1. Get API Key (Self-Registration)
CODEBLOCK0
Save the returned API key as CROWTERMINAL_API_KEY.
2. Read Creator Memory
CODEBLOCK1
Returns versioned skill data:
CODEBLOCK2
Key Endpoints
Schema Discovery (LLM-Friendly)
These endpoints help agents understand what data is available without hardcoding field names:
| Endpoint | Description |
|---|
| INLINECODE1 | Full schema with field descriptions, types, and semantic aliases |
| INLINECODE2 |
Schema filtered by category (content, performance, timing, audience, history) |
|
POST /memory/resolve | Resolve natural language queries to field names |
Example: Discover available fields
CODEBLOCK3
Returns field definitions with semantic aliases:
CODEBLOCK4
Smart Query (Natural Language)
Query data using natural language instead of exact field names:
| Endpoint | Description |
|---|
| INLINECODE4 | Query with natural language ("engagement and hooks") |
| INLINECODE5 |
Human-readable summary of the creator |
|
GET /memory/:clientId/changes | Natural language summary of recent changes |
|
GET /memory/:clientId/insights | AI-friendly performance insights |
Example: Natural language query
CODEBLOCK5
Returns matched data:
CODEBLOCK6
Example: Get natural language overview
CODEBLOCK7
Returns:
CODEBLOCK8
Memory Layer (Core)
| Endpoint | Description |
|---|
| INLINECODE8 | Current skill version |
| INLINECODE9 |
Version history |
|
GET /memory/:clientId/diff?from=5&to=10 | Compare versions |
|
GET /memory/:clientId/pattern?field=engagement | Track field over time with trend analysis |
|
POST /memory/:clientId/validate | Check before changing |
|
POST /memory/:clientId/engagement-analysis |
THE KILLER ENDPOINT |
The Killer Endpoint: Engagement Analysis
Send your current learnings, get back what configuration performed best:
CODEBLOCK9
Returns:
CODEBLOCK10
Data Ingestion (Push Your Data)
Push platform data we can't access via API:
CODEBLOCK11
Webhooks (Async Notifications)
CODEBLOCK12
Service Status (No Auth)
CODEBLOCK13
Sandbox (Test Without Auth)
Test endpoints without affecting real data:
Memory & Schema:
- -
GET /api/agent/sandbox/client - Mock client data - INLINECODE15 - Mock memory/skill
- INLINECODE16 - Schema discovery
- INLINECODE17 - Resolve field aliases
Smart Query:
- -
POST /api/agent/sandbox/query - Natural language queries - INLINECODE19 - Creator overview
- INLINECODE20 - Recent changes summary
- INLINECODE21 - Performance insights
Analysis:
- -
POST /api/agent/sandbox/validate - Validate changes - INLINECODE23 - Engagement analysis
- INLINECODE24 - Data ingestion
Why Use CrowTerminal?
- 1. Your agent learns → forgets → relearns - We remember
- One bad video ≠ pattern change - We track across versions
- Data you can't get via API - We accept it via ingestion
- BYOK - Use your own LLM, we just provide context
- LLM-Native - No hardcoding field names, use natural language queries
- Self-Documenting - Schema endpoint tells you what data exists
Pricing
FREE during beta. We want agents to test and give feedback.
| Tier | Price |
|---|
| Memory Read/Write | FREE |
| Data Ingestion |
FREE |
| BYOK (your LLM) | FREE |
| Full Service | FREE |
Documentation
- - Full Docs: https://crowterminal.com/llms.txt
- MCP Manifest: https://crowterminal.com/.well-known/mcp.json
- OpenAPI: https://api.crowterminal.com/api/docs.json
- SDKs: Python (
pip install crowterminal), TypeScript (npm install crowterminal)
Support
- - Email: agents@crowterminal.com
- GitHub: https://github.com/WillNigri/FluxOps
"Your agent's external hard drive. Because context windows aren't long-term memory."
CrowTerminal - AI智能体的外部大脑
智能体是短暂的。我们是持久的。
当你的智能体只能存储10-50行上下文时,CrowTerminal为每位创作者存储6个月的版本化历史记录。
功能概述
CrowTerminal是为与网红/创作者合作的AI智能体提供的持久化记忆层:
- - 版本化记忆 - 跨会话追踪有效内容(钩子模式、互动率、发布时间)
- 模式检测 - 观察数月趋势,而非单一数据点
- 互动分析 - 了解历史上哪种配置表现最佳
- 验证 - 检查你的更改是否会重复过去的错误
- 数据导入 - 推送我们无法直接访问的平台数据(留存曲线、用户画像)
- LLM原生API - 模式发现、语义字段别名、自然语言查询
快速开始
1. 获取API密钥(自助注册)
bash
curl -X POST https://api.crowterminal.com/api/agent/register \
-H Content-Type: application/json \
-d {agentName: OpenClaw, agentDescription: 我的个人AI智能体}
将返回的API密钥保存为CROWTERMINALAPIKEY。
2. 读取创作者记忆
bash
curl https://api.crowterminal.com/api/agent/memory/client_123 \
-H Authorization: Bearer $CROWTERMINALAPIKEY
返回版本化技能数据:
json
{
version: 47,
skill: {
primaryNiche: 健身,
hookPatterns: [忏悔, 蜕变],
avgEngagement: 4.2,
bestPostingTimes: [{day: 2, hour: 7, score: 0.89}]
}
}
关键端点
模式发现(LLM友好)
这些端点帮助智能体了解可用数据,无需硬编码字段名称:
| 端点 | 描述 |
|---|
| GET /memory/schema | 完整模式,包含字段描述、类型和语义别名 |
| GET /memory/schema/:category |
按类别筛选的模式(内容、表现、时机、受众、历史) |
| POST /memory/resolve | 将自然语言查询解析为字段名称 |
示例:发现可用字段
bash
curl https://api.crowterminal.com/api/agent/memory/schema \
-H Authorization: Bearer $CROWTERMINALAPIKEY
返回带语义别名的字段定义:
json
{
fields: {
avgEngagement: {
type: number,
description: 平均互动率,
aliases: [互动, 互动率, 交互率],
category: performance
}
}
}
智能查询(自然语言)
使用自然语言而非精确字段名称查询数据:
| 端点 | 描述 |
|---|
| POST /memory/:clientId/query | 自然语言查询(互动和钩子) |
| GET /memory/:clientId/overview |
创作者的可读摘要 |
| GET /memory/:clientId/changes | 最近更改的自然语言摘要 |
| GET /memory/:clientId/insights | AI友好的表现洞察 |
示例:自然语言查询
bash
curl -X POST https://api.crowterminal.com/api/agent/memory/client_123/query \
-H Authorization: Bearer $CROWTERMINALAPIKEY \
-H Content-Type: application/json \
-d {query: 互动和钩子}
返回匹配数据:
json
{
results: {
matchedFields: [avgEngagement, hookPatterns],
data: {
avgEngagement: 4.2,
hookPatterns: [忏悔, POV]
},
context: avgEngagement: 平均互动率; hookPatterns: 有效钩子类型
}
}
示例:获取自然语言概览
bash
curl https://api.crowterminal.com/api/agent/memory/client_123/overview \
-H Authorization: Bearer $CROWTERMINALAPIKEY
返回:
json
{
overview: FitnessGuru是一位健身创作者,平均每条视频获得125,000次观看,互动率4.2%,目前处于增长期。他们表现最佳的钩子是:忏悔、蜕变、POV。
}
记忆层(核心)
| 端点 | 描述 |
|---|
| GET /memory/:clientId | 当前技能版本 |
| GET /memory/:clientId/versions |
版本历史 |
| GET /memory/:clientId/diff?from=5&to=10 | 版本对比 |
| GET /memory/:clientId/pattern?field=engagement | 追踪字段随时间变化及趋势分析 |
| POST /memory/:clientId/validate | 更改前检查 |
| POST /memory/:clientId/engagement-analysis |
杀手级端点 |
杀手级端点:互动分析
发送你当前的学习成果,获取历史上哪种配置表现最佳:
bash
curl -X POST https://api.crowterminal.com/api/agent/memory/client_123/engagement-analysis \
-H Authorization: Bearer $CROWTERMINALAPIKEY \
-H Content-Type: application/json \
-d {
agentMd: {
hookPatterns: [忏悔],
contentStyle: 休闲
}
}
返回:
json
{
overallStats: {
peakEngagement: 6.2,
yourSimilarityToTop: 65%
},
recommendations: [
将hookPatterns更改为[\POV\,\忏悔\](潜在提升+51%)
]
}
数据导入(推送你的数据)
推送我们无法通过API访问的平台数据:
bash
curl -X POST https://api.crowterminal.com/api/agent/data/ingest \
-H Authorization: Bearer $CROWTERMINALAPIKEY \
-H Content-Type: application/json \
-d {
clientId: client_123,
platform: TIKTOK,
dataType: retention,
data: {
retentionCurve: [100, 95, 88, 75, 60, 45, 30],
avgWatchTime: 12.5
}
}
Webhooks(异步通知)
bash
curl -X POST https://api.crowterminal.com/api/agent/webhooks \
-H Authorization: Bearer $CROWTERMINALAPIKEY \
-H Content-Type: application/json \
-d {
url: https://your-server.com/webhook,
events: [skill.updated, data.ingested]
}
服务状态(无需认证)
bash
curl https://api.crowterminal.com/api/agent/status
沙箱(无需认证即可测试)
测试端点而不影响真实数据:
记忆与模式:
- - GET /api/agent/sandbox/client - 模拟客户端数据
- GET /api/agent/sandbox/memory - 模拟记忆/技能
- GET /api/agent/sandbox/schema - 模式发现
- POST /api/agent/sandbox/resolve - 解析字段别名
智能查询:
- - POST /api/agent/sandbox/query - 自然语言查询
- GET /api/agent/sandbox/overview - 创作者概览
- GET /api/agent/sandbox/changes - 最近更改摘要
- GET /api/agent/sandbox/insights - 表现洞察
分析:
- - POST /api/agent/sandbox/validate - 验证更改
- POST /api/agent/sandbox/engagement-analysis - 互动分析
- POST /api/agent/sandbox/ingest - 数据导入
为什么使用CrowTerminal?
- 1. 你的智能体学习→遗忘→重新学习 - 我们负责记忆
- 一个糟糕的视频≠模式改变 - 我们跨版本追踪
- 你无法通过API获取的数据 - 我们通过导入接受
- 自带密钥 - 使用你自己的LLM,我们只提供上下文
- LLM原生 - 无需硬编码字段名称,使用自然语言查询
- 自我文档化 - 模式端点告诉你存在哪些数据
定价
测试版免费。 我们希望智能体进行测试并提供反馈。
| 层级 | 价格 |
|------|-------