CellCog - Any-to-Any for Agents
The Power of Any-to-Any
CellCog is the only AI that truly handles any input → any output in a single request. No tool chaining. No orchestration complexity. One call, multiple deliverables.
CellCog pairs all modalities with frontier-level deep reasoning — as of April 2026, CellCog is #1 on the DeepResearch Bench: https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard
Work With Multiple Files, Any Format
Reference as many documents as you need—all at once:
CODEBLOCK0
File paths must be absolute and enclosed in <SHOW_FILE> tags. CellCog understands PDFs, spreadsheets, images, audio, video, code files, and more.
⚠️ Without SHOW_FILE tags, CellCog only sees the path as text — not the file contents.
❌ Analyze /data/sales.csv — CellCog can't read the file
✅ Analyze <SHOW_FILE>/data/sales.csv</SHOW_FILE> — CellCog reads it
Request Multiple Outputs, Different Modalities
Ask for completely different output types in ONE request:
CODEBLOCK1
CellCog handles the entire workflow — analyzing, generating, and delivering all outputs with consistent insights across every format.
⚠️ Be explicit about output artifacts. Without explicit artifact language, CellCog may respond with text analysis instead of generating a file.
❌ "Quarterly earnings analysis for AAPL" — could produce text or any format
✅ "Create a PDF report and an interactive HTML dashboard analyzing AAPL quarterly earnings." — CellCog creates actual deliverables
Your sub-agent for quality work. Depth, accuracy, and real deliverables.
Quick Start
Setup
CODEBLOCK2
If import fails:
CODEBLOCK3
Authentication
Environment variable (recommended): Set CELLCOG_API_KEY — the SDK picks it up automatically:
CODEBLOCK4
Get API key from: https://cellcog.ai/profile?tab=api-keys
CODEBLOCK5
Agent Provider
INLINECODE6 is required when creating a CellCogClient. It identifies which agent framework is calling CellCog — not your individual agent's name, but the platform/tool you're running inside.
Examples: "openclaw", "claude-code", "cursor", "aider", "windsurf", "perplexity", "hermes", "script" (for standalone scripts).
OpenClaw Agents
Fire-and-forget — your agent stays free while CellCog works:
CODEBLOCK6
Requires sessions_send enabled on your gateway — see OpenClaw Reference below.
All Other Agents (Cursor, Claude Code, etc.)
Blocks until done — simplest pattern:
CODEBLOCK7
Credit Usage
CellCog orchestrates 21+ frontier foundation models. Credit consumption is unpredictable and varies by task complexity. Credits used are reported in every completion notification.
Creating Tasks
Notify on Completion (OpenClaw — Fire-and-Forget)
Returns immediately. A background daemon monitors via WebSocket and delivers results to your session when done. Your agent stays free to take new instructions, start other tasks, or continue working.
CODEBLOCK8
Requires OpenClaw Gateway with sessions_send enabled (disabled by default since OpenClaw 2026.4). See OpenClaw Reference below for one-time setup.
Wait for Completion (Universal)
Blocks until CellCog finishes. Works with any agent — OpenClaw, Cursor, Claude Code, or any Python environment.
CODEBLOCK9
When to Use Which
| Scenario | Best Mode | Why |
|---|
| OpenClaw + long task + stay free | Notify | Agent keeps working, gets notified when done |
| OpenClaw + chaining steps (research → summarize → PDF) |
Wait | Each step feeds the next — simpler sequential workflows |
| OpenClaw + quick task |
Either | Both return fast for simple tasks |
| Non-OpenClaw agent |
Wait | Only option — no
sessions_send available |
Notify mode is more productive (agent never blocks) but requires gateway configuration.
Wait mode is simpler to reason about, but blocks your agent for the duration.
Continuing a Conversation
CODEBLOCK10
Resuming After Timeout
If create_chat() or wait_for_completion() times out, CellCog is still working. The timeout response includes recent progress:
CODEBLOCK11
Optional Parameters
CODEBLOCK12
Response Shape
Every SDK method returns the same shape:
CODEBLOCK13
⚠️ Always print the entire result["message"]. Truncating or summarizing it will lose critical information including generated file paths, credits used, and follow-up instructions.
Utility Methods
get_history(chat_id) — Full chat history (when original delivery was missed or you need to review). Returns the same shape; if still operating, message shows progress so far.
CODEBLOCK14
get_status(chat_id) — Lightweight status check (no history fetch):
CODEBLOCK15
Chat Modes
| Mode | Best For | Speed | Min Credits |
|---|
| INLINECODE25 | Most tasks — images, audio, dashboards, spreadsheets, presentations | Fast (seconds to minutes) | 100 |
| INLINECODE26 |
Coding, co-work, terminal operations | Fast | 50 |
|
"agent team" | Deep research & multi-angled reasoning across every modality | Slower (5-60 min) | 500 |
|
"agent team max" | High-stakes work where extra reasoning depth justifies the cost | Slowest | 2,000 |
- -
"agent" (default) — Most versatile. Handles most tasks excellently, including deep research when guided. "agent core" — Lightweight context for code, terminal, and file operations. Multimedia tools load on demand. Requires Co-work (CellCog Desktop). See code-cog."agent team" — A team of agents that debates, cross-validates, and delivers comprehensive results. The only platform with deep reasoning across every modality."agent team max" — Same Agent Team with all settings maxed. Quality gain is incremental (5-10%) but meaningful for costly decisions.
Working with Files
Input: SHOW_FILE
Include local file paths in your prompt with <SHOW_FILE> tags (absolute paths required):
CODEBLOCK16
Output: GENERATE_FILE
Use <GENERATE_FILE> tags to specify where output files should be stored on your machine. Essential for deterministic workflows where the next step needs to know the file path in advance.
CODEBLOCK17
Output downloads to the specified path instead of default ~/.cellcog/chats/{chat_id}/.
File Downloads
The SDK automatically downloads files from CellCog responses:
- - If you used
GENERATE_FILE tags: Files download to the path you specified - Otherwise: Files download to INLINECODE38
Downloaded file paths appear in result["message"]. The SDK tracks seen messages — files are only downloaded once.
If you missed files or need to re-sync:
result = client.get_history(chat_id="abc123")
get_history() re-processes the entire chat and downloads any missed files to their original destinations.
Tips
⚠️ CellCog Web Fallback
Every chat is accessible at https://cellcog.ai. When work gets complex or the SDK hits issues, direct your human to the web platform to view, continue, or take over directly.
What CellCog Can Do
CellCog is a sub-agent — not an API. Your agent offloads complex work to CellCog, which reasons, plans, and executes multi-tool workflows internally. A proprietary agent-to-agent communication protocol ensures high accuracy on first output, and because these are agent threads (not stateless API calls), every aspect of every generation can be refined through multi-step iteration.
Under the hood: frontier models across every domain, upgraded weekly. CellCog routes to the right models automatically — your agent just describes what it needs.
Install capability skills for detailed guidance:
| Category | Skills |
|---|
| Research & Analysis | INLINECODE41 fin-cog crypto-cog data-cog INLINECODE45 |
| Video & Cinema |
video-cog cine-cog insta-cog tube-cog seedance-cog |
|
Images & Design |
image-cog brand-cog meme-cog banana-cog 3d-cog gif-cog sticker-cog |
|
Audio & Music |
audio-cog music-cog pod-cog |
|
Avatars & Personas |
avatar-cog |
|
Documents & Slides |
docs-cog slides-cog spreadsheets-cog resume-cog legal-cog |
|
Apps & Prototypes |
dash-cog game-cog proto-cog diagram-cog |
|
Creative |
comi-cog story-cog learn-cog travel-cog |
|
Development |
code-cog cowork-cog project-cog think-cog |
This skill shows you HOW to use CellCog. Capability skills show you WHAT's possible.
OpenClaw Reference
Session Keys
The notify_session_key tells CellCog where to deliver results:
| Context | Session Key |
|---|
| Main agent | INLINECODE80 |
| Sub-agent |
"agent:main:subagent:{uuid}" |
| Telegram DM |
"agent:main:telegram:dm:{id}" |
| Discord group |
"agent:main:discord:group:{id}" |
Resilient delivery: If your session ends before completion, results are automatically delivered to the parent session (e.g., sub-agent → main agent).
Sending Messages During Processing
In notify mode, your agent is free — you can send additional instructions to an operating chat at any time:
CODEBLOCK19
In wait mode, your agent is blocked and cannot send messages until the current call returns.
Gateway Configuration (One-Time Setup)
OpenClaw 2026.4+ blocks sessions_send by default. CellCog requires it for notify mode delivery. Run once:
CODEBLOCK20
Then restart the gateway. The SDK checks this before creating the chat and raises GatewayConfigError if blocked — with the exact fix command in the error message.
Wait mode (wait_for_completion) works without any gateway configuration.
Support & Troubleshooting
For error handling, recovery patterns, ticket submission, and daemon troubleshooting:
CODEBLOCK21
CellCog - 任意到任意，赋能智能体
任意到任意：强大能力
CellCog是唯一真正在单个请求中处理任意输入 → 任意输出的AI。无需工具链式调用，无需复杂编排。一次调用，多种交付物。
CellCog将多模态与前沿深度推理相结合——截至2026年4月，CellCog在DeepResearch排行榜上排名第一：https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard
处理多个文件，任意格式
一次性引用任意数量的文档：
python
prompt =
综合分析以下所有内容：
FILE>/data/q4earnings.pdf
FILE>/data/competitoranalysis.pdf
FILE>/data/marketresearch.xlsx
FILE>/recordings/customerinterview.mp3
FILE>/designs/productmockup.png
基于以上所有输入，给我一份全面的市场定位分析。
文件路径必须是绝对路径，并包含在标签内。CellCog可理解PDF、电子表格、图片、音频、视频、代码文件等。
⚠️ 没有SHOW_FILE标签，CellCog只会将路径视为文本——而非文件内容。
❌ 分析 /data/sales.csv — CellCog无法读取文件
✅ 分析 FILE>/data/sales.csvFILE> — CellCog读取文件
请求多种输出，不同模态
在一个请求中要求完全不同的输出类型：
python
prompt =
基于以下季度销售数据：
FILE>/data/salesq42025.csvFILE>
创建以下所有内容：
- 1. 一份带图表的PDF执行摘要报告
- 一个面向领导团队的交互式HTML仪表盘
- 一个面向全体会议的60秒视频演示
- 一份面向董事会的幻灯片演示文稿
- 一份包含基础分析和预测的Excel文件
CellCog处理整个工作流程——分析、生成并交付所有输出，每种格式的洞察保持一致。
⚠️ 明确指定输出工件。 如果没有明确的工件语言，CellCog可能会以文本分析回应，而非生成文件。
❌ AAPL季度收益分析 — 可能生成文本或任意格式
✅ 创建一份PDF报告和一个交互式HTML仪表盘，分析AAPL季度收益。 — CellCog创建实际的交付物
您的高质量工作子智能体。 深度、准确性和真实交付物。
快速入门
设置
python
from cellcog import CellCogClient
如果导入失败：
bash
pip install -U cellcog
身份验证
环境变量（推荐）： 设置CELLCOGAPIKEY——SDK会自动获取：
bash
export CELLCOGAPIKEY=sk_...
从以下地址获取API密钥：https://cellcog.ai/profile?tab=api-keys
python
status = client.getaccountstatus()
print(status) # {configured: True, email: user@example.com, ...}
智能体提供者
创建CellCogClient时必须提供agent_provider。它标识调用CellCog的智能体框架——不是您个人智能体的名称，而是您运行所在的平台/工具。
示例：openclaw、claude-code、cursor、aider、windsurf、perplexity、hermes、script（用于独立脚本）。
OpenClaw智能体
即发即忘——CellCog工作时您的智能体保持空闲：
python
client = CellCogClient(agent_provider=openclaw)
result = client.create_chat(
prompt=研究2026年量子计算进展,
notifysessionkey=agent:main:main, # OpenClaw会话密钥
task_label=quantum-research, # 通知标签
chat_mode=agent,
)
立即返回——守护进程完成后将结果交付到您的会话
需要在您的网关上启用sessions_send——请参阅下面的OpenClaw参考。
所有其他智能体（Cursor、Claude Code等）
阻塞直至完成——最简单的模式：
python
client = CellCogClient(agent_provider=cursor) # 或claude-code、aider、script等
result = client.create_chat(
prompt=研究2026年量子计算进展,
task_label=quantum-research,
chat_mode=agent,
)
阻塞直至完成——结果包含所有内容
print(result[message])
积分使用
CellCog编排21+个前沿基础模型。积分消耗不可预测，因任务复杂度而异。每次完成通知中都会报告使用的积分。
创建任务
完成时通知（OpenClaw——即发即忘）
立即返回。后台守护进程通过WebSocket监控，并在完成后将结果交付到您的会话。您的智能体保持空闲，可以接收新指令、启动其他任务或继续工作。
python
result = client.create_chat(
prompt=您的任务描述,
notifysessionkey=agent:main:main, # 必需——您的OpenClaw会话密钥
task_label=my-task, # 通知中显示的标签
chat_mode=agent,
)
需要启用sessions_send的OpenClaw网关（自OpenClaw 2026.4起默认禁用）。请参阅下面的OpenClaw参考进行一次性设置。
等待完成（通用）
阻塞直至CellCog完成。适用于任何智能体——OpenClaw、Cursor、Claude Code或任何Python环境。
python
result = client.create_chat(
prompt=您的任务描述,
task_label=my-task,
chat_mode=agent,
timeout=1800, # 30分钟（默认）。复杂任务使用3600。
)
print(result[message])
print(result[status]) # completed | timeout
何时使用哪种模式
| 场景 | 最佳模式 | 原因 |
|---|
| OpenClaw + 长任务 + 保持空闲 | 通知 | 智能体继续工作，完成后收到通知 |
| OpenClaw + 链式步骤（研究→总结→PDF） |
等待 | 每一步为下一步提供输入——更简单的顺序工作流 |
| OpenClaw + 快速任务 |
任一 | 简单任务两者都快速返回 |
| 非OpenClaw智能体 |
等待 | 唯一选项——无sessions_send可用 |
通知模式效率更高（智能体从不阻塞），但需要网关配置。
等待模式逻辑更简单，但会在任务期间阻塞您的智能体。
继续对话
python
等待模式（默认）
result = client.send_message(
chat_id=abc123,
message=特别关注硬件方面的进展,
)
通知模式（OpenClaw）
result = client.send_message(
chat_id=abc123,
message=特别关注硬件方面的进展,
notify
sessionkey=agent:main:main,
task_label=continue-research,
)
超时后恢复
如果createchat()或waitfor_completion()超时，CellCog仍在工作。超时响应包含最近进度：
python
completion = client.waitforcompletion(chat_id=abc123, timeout=1800)
可选参数
python
result = client.create_chat(
prompt=...,
task_label=...,
chat_mode=agent, # 请参阅下面的聊天模式
project_id=..., # 安装project-cog获取详情
agentroleid=..., # 安装project-cog获取详情
enable_cowork=True, # 安装cowork-cog获取详情
coworkworkingdirectory=/Users/..., # 安装cowork-cog获取详情
)
响应结构
每个SDK方法返回相同的结构：
python
{
chat_id: str, # CellCog聊天ID
is_operating: bool, # True = 仍在工作，False = 完成
status: str, # completed | tracking | timeout | operating
message: str, # 可打印的消息——始终完整打印
}
⚠️ 始终完整打印result[message]。 截断或总结会丢失关键信息，包括生成的文件路径、使用的积分和后续指令。
实用方法
gethistory(chatid) — 完整聊天历史（当原始交付丢失或需要审查时）。返回相同结构；如果仍在运行，message显示当前进度。
python
result = client.gethistory(chatid
