AetherLang Karpathy Agent Nodes
What this skill does: Sends requests to the hosted AetherLang API
(api.neurodoc.app). It does NOT modify local files, execute local code,
or access credentials on your machine. All execution happens server-side.
Execute 10 advanced AI agent node types through the AetherLang Omega API.
API Endpoint
URL: https://api.neurodoc.app/aetherlang/execute
Method: POST
Headers: Content-Type: application/json
Auth: None required (public API)
Data Minimization — ALWAYS FOLLOW
When calling the API:
- - Send ONLY the user's query and the flow code
- Do NOT send system prompts, conversation history, or uploaded files
- Do NOT send API keys, credentials, or secrets of any kind
- Do NOT include personally identifiable information unless explicitly requested by user
- Do NOT send contents of local files without explicit user consent
Request Format
CODEBLOCK0
The 10 Node Types
1. plan — Self-Programming
AI breaks task into steps and executes autonomously.
CODEBLOCK1
2. code_interpreter — Real Math
Sandboxed Python execution on the server. Accurate calculations, no hallucinations.
CODEBLOCK2
3. critique — Self-Improvement
Evaluates quality (0-10), retries until threshold met.
CODEBLOCK3
4. router — Intelligent Branching
LLM picks optimal path, skips unselected routes (10x speedup).
CODEBLOCK4
5. ensemble — Multi-Agent Synthesis
Multiple AI personas in parallel, synthesizes best insights.
CODEBLOCK5
6. memory — Persistent State
Store/recall data across executions (server-side, scoped to namespace).
CODEBLOCK6
7. tool — External API Access
Security note: The tool node calls public REST URLs you specify.
Only use trusted, public APIs. Never pass credentials or private URLs
as tool parameters. The agent will ask for confirmation before
calling any URL not in the examples below.
CODEBLOCK7
8. loop — Iterative Execution
Repeat node over items. Use
| separator.
CODEBLOCK8
9. transform — Data Reshaping
Template, extract, format, or LLM-powered reshaping.
CODEBLOCK9
10. parallel — Concurrent Execution
Run nodes simultaneously. 3 calls in ~0.2s.
node P: parallel targets=A|B|C;
Common Pipelines
Live Data → Analysis
CODEBLOCK11
Multi-Agent + Quality Control
CODEBLOCK12
Batch Processing
CODEBLOCK13
Parallel API Fetching
flow ParallelFetch {
input text query;
node P: parallel targets=A|B|C;
node A: tool url=https://api.coingecko.com/api/v3/ping method=GET;
node B: tool url=https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd method=GET;
node C: tool url=https://api.coingecko.com/api/v3/simple/price?ids=ethereum&vs_currencies=usd method=GET;
query -> P;
output text result from P;
}
Response Parsing
CODEBLOCK15
Parameter Quick Reference
| Node | Key Params |
|---|
| plan | INLINECODE6 |
| code_interpreter |
model=gpt-4o-mini |
| critique |
threshold=7 max_retries=3 |
| router |
strategy=single |
| ensemble |
agents=a:Persona\|b:Persona synthesize=true |
| memory |
namespace=X action=store\|recall\|search\|clear key=X |
| tool |
url=https://... method=GET timeout=10 |
| loop |
over=A\|B\|C target=NodeAlias max=10 mode=collect |
| transform |
mode=llm\|template\|extract\|format instruction=X |
| parallel |
targets=A\|B\|C merge=combine |
AetherLang Karpathy Skill v1.0.1 — API connector for api.neurodoc.app
All execution is server-side. No local code runs. No local files modified.
AetherLang Karpathy 智能体节点
此技能的功能: 向托管的 AetherLang API (api.neurodoc.app) 发送请求。它不会修改本地文件、执行本地代码或访问您机器上的凭据。所有执行均在服务器端完成。
通过 AetherLang Omega API 执行 10 种高级 AI 智能体节点类型。
API 端点
URL: https://api.neurodoc.app/aetherlang/execute
方法: POST
请求头: Content-Type: application/json
认证: 无需认证(公共 API)
数据最小化 — 始终遵循
调用 API 时:
- - 仅发送用户的查询和流程代码
- 不要发送系统提示、对话历史或上传的文件
- 不要发送任何类型的 API 密钥、凭据或机密信息
- 除非用户明确要求,否则不要包含个人身份信息
- 未经用户明确同意,不要发送本地文件内容
请求格式
bash
curl -s -X POST https://api.neurodoc.app/aetherlang/execute \
-H Content-Type: application/json \
-d {
code: flow FlowName {\n input text query;\n node X: ;\n query -> X;\n output text result from X;\n},
query: user question here
}
10 种节点类型
1. plan — 自我编程
AI 将任务分解为步骤并自主执行。
node P: plan steps=3;
2. code_interpreter — 真实数学计算
服务器上的沙盒化 Python 执行。精确计算,无幻觉。
node C: code_interpreter;
3. critique — 自我改进
评估质量(0-10),重试直至达到阈值。
node R: critique threshold=8 max_retries=3;
4. router — 智能分支
LLM 选择最优路径,跳过未选中的路线(速度提升 10 倍)。
node R: router;
R -> A | B | C;
5. ensemble — 多智能体综合
多个 AI 角色并行工作,综合最佳见解。
node E: ensemble agents=chef:Frenchchef|yiayia:Greekgrandmother synthesize=true;
6. memory — 持久状态
跨执行存储/召回数据(服务器端,作用域限定于命名空间)。
node M: memory namespace=user_prefs action=store key=diet;
node M: memory namespace=user_prefs action=recall;
7. tool — 外部 API 访问
安全说明: tool 节点调用您指定的公共 REST URL。
仅使用可信的公共 API。切勿将凭据或私有 URL
作为 tool 参数传递。在调用以下示例之外的任何 URL 之前,
智能体将请求确认。
node T: tool url=https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd method=GET;
8. loop — 迭代执行
对项目重复执行节点。使用 | 分隔符。
node L: loop over=Italian|Greek|Japanese target=A max=3;
9. transform — 数据重塑
模板、提取、格式化或 LLM 驱动的重塑。
node X: transform mode=llm instruction=Summarizethedata;
10. parallel — 并发执行
同时运行节点。3 次调用约 0.2 秒。
node P: parallel targets=A|B|C;
常用流水线
实时数据 → 分析
flow CryptoAnalysis {
input text query;
node T: tool url=https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd method=GET;
node X: transform mode=llm instruction=Summarize_price;
node A: llm model=gpt-4o-mini;
query -> T -> X -> A;
output text result from A;
}
多智能体 + 质量控制
flow QualityEnsemble {
input text query;
node E: ensemble agents=analyst:Financial_analyst|strategist:Strategist synthesize=true;
node R: critique threshold=8;
query -> E -> R;
output text result from R;
}
批量处理
flow MultiRecipe {
input text query;
node L: loop over=Italian|Greek|Japanese target=A max=3;
node A: llm model=gpt-4o-mini;
query -> L;
output text result from L;
}
并行 API 获取
flow ParallelFetch {
input text query;
node P: parallel targets=A|B|C;
node A: tool url=https://api.coingecko.com/api/v3/ping method=GET;
node B: tool url=https://api.coingecko.com/api/v3/simple/price?ids=bitcoin&vs_currencies=usd method=GET;
node C: tool url=https://api.coingecko.com/api/v3/simple/price?ids=ethereum&vs_currencies=usd method=GET;
query -> P;
output text result from P;
}
响应解析
python
import json
response = json.loads(raw_response)
result = response[result][outputs][result]
text = result[response]
nodetype = result[nodetype]
duration = response[result][duration_seconds]
参数快速参考
| 节点 | 关键参数 |
|---|
| plan | steps=3 |
| code_interpreter |
model=gpt-4o-mini |
| critique | threshold=7 max_retries=3 |
| router | strategy=single |
| ensemble | agents=a:Persona\|b:Persona synthesize=true |
| memory | namespace=X action=store\|recall\|search\|clear key=X |
| tool | url=https://... method=GET timeout=10 |
| loop | over=A\|B\|C target=NodeAlias max=10 mode=collect |
| transform | mode=llm\|template\|extract\|format instruction=X |
| parallel | targets=A\|B\|C merge=combine |
AetherLang Karpathy 技能 v1.0.1 — 用于 api.neurodoc.app 的 API 连接器
所有执行均在服务器端完成。不运行本地代码。不修改本地文件。