Step 1
Choose an agent client
Pick a client that supports custom model providers, compatible APIs or MCP. Begin with local, reversible tasks.
View agent options →NEW SESSIONWhat would you like to build?
AGENT
Start with DeepSeek.
Build an AI Agent.
Learn how agents plan tasks, use tools and keep working toward a goal.
Connect the DeepSeek API to the coding and automation clients you already use.
This is an independent DeepSeek Agent guide. It is not affiliated with or endorsed by DeepSeek.
⌄01
DEEPSEEK AGENT BASICS
What is a DeepSeek Agent?
It is not a new model. It is a way for a model to understand its environment, use tools and keep working until a task is complete.
A standard chat model mainly produces a single response. An agent works in a loop: observe the current state, make a plan, choose tools, inspect results and decide what to do next.
Models, tools, skills, sessions, sandboxes and storage can be combined for each task. Important actions should still require human approval.
Terminal — agent
$ deepseek-agent start✓ Model connected✓ Tools ready› Describe your first task...QUICK START
How to use a DeepSeek Agent
Connect the model first, then grant tools and workspace access gradually.
Step 2
Configure the model and API key
Enter the API key, base URL and model ID in your client's Models or Provider settings. Never commit secrets to a public repository.
View API setup →Step 3
Select a workspace and run a task
State the goal, source material, restrictions and desired output. Keep permissions minimal and review logs and artifacts after execution.
Explore workflow tools →✓ Goal understood✓ Workspace inspected↻ Running and validating○ Waiting for approval
EXTEND CAPABILITIES
Tools and workflows
Turn an agent from something that answers into something that completes work.
02
CONNECT THE MODEL
DeepSeek API integration
Clients use different protocols. Confirm the API format your client supports before entering configuration.
| Item | Compatible chat API | Tool use | Agent client |
|---|---|---|---|
| Core input | messages | Model request + tool result | Goal + workspace |
| State | Client maintains history | Tool results return to loop | Plans, steps and artifacts |
| Best for | Chat and content | Search, code and data | Real multi-step tasks |
| Safety | Protect the API key | Restrict tools and arguments | Human approval for risky actions |
Model connection
Verify one standard conversation before enabling tools.
https://api.deepseek.comTool connection
Add search, code, files or MCP tools one at a time.
WEB · CODE · FILES · MCPApproval boundary
Require approval for publishing, payments, deletion, permissions and private data.
AGENT → REVIEW → ACTIONCHOOSE YOUR WORKFLOW
Agent use cases with DeepSeek
Start with bounded tasks whose results are easy to verify.
Coding agent
Read code, edit files, run tests and deliver a reviewable diff.
View guideResearch agent
Collect sources, cross-check claims and create a cited report.
View guideData agent
Clean data, identify patterns and prepare decision-ready analysis.
View guideCreative agent
Turn briefs into copy, visual directions, images and videos.
Create free ↗PIXPIX · FREE AI CREATIVE TOOLSTurn agent ideas
Turn agent ideas
into images and videos.
Create AI images and videos from a simple description for marketing, product visuals, social content and creative proposals.
Create free with PixPix ↗AI IMAGE
AI VIDEO
✦
REAL WORKFLOWS
A day with AI agents
Terminal · Coding
“This test is failing. Find the cause and fix it.”
The agent reads the stack trace, inspects related code, applies a fix and runs verification before summarizing changes.
Browser · Research
“Compare these three products and recommend one.”
The agent gathers sources, verifies specifications and structures the findings in a clear comparison.
Workspace · Creative
“Turn today's campaign plan into social assets.”
The agent extracts key messages and visual directions, then opens creative tools for image and video drafts.
FAQ
DeepSeek Agent
questions
Common questions about APIs, tools, permissions and workflows.
What is a DeepSeek AI Agent?+
It is an agent workflow powered by a DeepSeek model that plans steps, calls tools, checks results and keeps working toward a goal.
How is an agent different from a chatbot?+
A chatbot mainly returns a response. An agent maintains task state and carries out multi-step work across search, code, files and data.
Do I need to know programming?+
Not always. Many clients offer visual configuration, while advanced tool integrations and deployments usually require technical knowledge.
How should I protect my API key?+
Use environment variables or secure client storage. Never expose keys in public repositories, screenshots, logs or frontend code.
Can an agent run fully autonomously?+
Low-risk, reversible and verifiable tasks can be automated. Money, private data, publishing, deletion and permission changes should require approval.
Is this an official DeepSeek website?+
No. This is an independent guide and is not affiliated with or endorsed by DeepSeek.
Ready to use
DeepSeek Agent?
Begin with one clear task, then connect models, tools and workspaces gradually.
View quick startBrowse agent use cases