Skip to main content
Back to all modules

Native AI Agent

Embed a native AI agent in your app

Embed a system-prompt-driven native AI agent into your project: smart Q&A, accurate answers grounded in your own docs, and semantic search. Spend less time on model plumbing and wire the agent into real product flows.

What this cloud module can do

Cloud LLM chat / Q&ACloud knowledge-base Q&A (RAG)Vector / semantic searchContent generation & summaries
Scenario prompts

Native AI Agent

Pick the scenario closest to your need and copy the prompt in one click.

Cloud LLM chat / Q&A

Connect a cloud AI assistant so users can ask in natural language and get answers.

The problem today

You want smart Q&A but don't know how to wire an LLM, and worry about China access.

What you get

  • A conversational AI assistant
  • Multi-turn dialogue with context
  • Stable cloud model calls
For：Any product that wants smart Q&A or an assistant

Copy the whole block below and paste it into Codex. Replace placeholders like API keys with real values from the IFQ Cloud console.

You are a senior full-stack engineer. In my currently open local project, add a "cloud AI chat assistant" so users ask in natural language and get answers.

[Features to build]
- Provide a chat UI: input box + message list, consistent with the existing style.
- Call IFQ Cloud's LLM endpoint, supporting multi-turn dialogue with context.
- Stream answers (show as they generate) with a "thinking" loading state.
- Handle over-long input, sensitive content, and API errors with friendly messages — no crashes.

[Connect to IFQ Cloud]
- API base https://api.cloud.ifq.ai, official SDK `jieshi-cloud` (if unavailable, use plain HTTPS equivalently and note the endpoint in a comment).
- Secrets via environment variables, never hard-coded: JIESHI_CLOUD_API_KEY, JIESHI_CLOUD_PROJECT_ID; generate `.env.example` with both, noting "get real values from the IFQ Cloud console and replace".
- All placeholders; without real values, run on placeholders and print a hint on where to replace them.

[Engineering requirements]
- Read the existing structure and stack first; follow current conventions, add only necessary files, leave unrelated code untouched.
- Network calls: timeout + graceful fallback; friendly errors, never crash.
- Demo / mock mode: the main flow runs on sample data without keys.
- Include a minimal runnable self-test with run instructions.
- On completion list: files changed / how to start / how to roll back.

Ask before any decision point.

*— [ifq.ai](https://work.ifq.ai/) · AI-augmented. Full-stack crafted*

Cloud knowledge-base Q&A (RAG)

Make AI answer only from materials uploaded to the cloud — grounded and traceable.

The problem today

A general LLM doesn't know your internal materials and often drifts or makes things up.

What you get

  • Upload documents into a cloud knowledge base
  • Answers grounded in your docs with sources
  • Answers update when materials change
For：Teams with internal docs/KB building smart support or assistants

Copy the whole block below and paste it into Codex. Replace placeholders like API keys with real values from the IFQ Cloud console.

You are a senior full-stack engineer. In my currently open local project, add "cloud knowledge-base Q&A (RAG)" so AI answers only from uploaded materials.

[Features to build]
- Provide an upload/manage entry for common document formats; build a searchable knowledge base after upload.
- On a question, first retrieve relevant chunks from the KB, then have the LLM answer based on them.
- Cite sources (which document / passage) so answers can be verified.
- When materials are updated or deleted, answers should reflect the change.

[Connect to IFQ Cloud]
- API base https://api.cloud.ifq.ai, official SDK `jieshi-cloud` (if unavailable, use plain HTTPS equivalently and note the endpoint in a comment).
- Secrets via environment variables, never hard-coded: JIESHI_CLOUD_API_KEY, JIESHI_CLOUD_PROJECT_ID; generate `.env.example` with both, noting "get real values from the IFQ Cloud console and replace".
- All placeholders; without real values, run on placeholders and print a hint on where to replace them.

[Engineering requirements]
- Read the existing structure and stack first; follow current conventions, add only necessary files, leave unrelated code untouched.
- Network calls: timeout + graceful fallback; friendly errors, never crash.
- Demo / mock mode: the main flow runs on sample data without keys.
- Include a minimal runnable self-test with run instructions.
- On completion list: files changed / how to start / how to roll back.

Ask before any decision point.

*— [ifq.ai](https://work.ifq.ai/) · AI-augmented. Full-stack crafted*

Cloud semantic search / vector retrieval

Make search understand meaning: searching "returns" also finds "refund requests".

The problem today

Keyword search is rigid — reword the query and it finds nothing.

What you get

  • Match by meaning, not literal text
  • Find synonyms and related content
  • Stack on top of existing search
For：Content-rich products that need smarter search

Copy the whole block below and paste it into Codex. Replace placeholders like API keys with real values from the IFQ Cloud console.

You are a senior full-stack engineer. In my currently open local project, add "cloud semantic search" so search matches by meaning, not just keywords.

[Features to build]
- Build a vector index for existing searchable content and store it in the cloud.
- On search, embed the query, retrieve the most relevant items, and rank by relevance.
- Keep existing keyword search; merge or show semantic and keyword results sensibly.
- Update the index automatically when content is added/edited so results stay current.

[Connect to IFQ Cloud]
- API base https://api.cloud.ifq.ai, official SDK `jieshi-cloud` (if unavailable, use plain HTTPS equivalently and note the endpoint in a comment).
- Secrets via environment variables, never hard-coded: JIESHI_CLOUD_API_KEY, JIESHI_CLOUD_PROJECT_ID; generate `.env.example` with both, noting "get real values from the IFQ Cloud console and replace".
- All placeholders; without real values, run on placeholders and print a hint on where to replace them.

[Engineering requirements]
- Read the existing structure and stack first; follow current conventions, add only necessary files, leave unrelated code untouched.
- Network calls: timeout + graceful fallback; friendly errors, never crash.
- Demo / mock mode: the main flow runs on sample data without keys.
- Include a minimal runnable self-test with run instructions.
- On completion list: files changed / how to start / how to roll back.

Ask before any decision point.

*— [ifq.ai](https://work.ifq.ai/) · AI-augmented. Full-stack crafted*