What can I do for you?
Hand your task to Pocket Expert AI — it plans, investigates, and executes on its own, delivering research reports, slide decks, and working apps.
One sentence in, real deliverables out
It plans autonomously and drives tools through every step of the investigation, with execution that's transparent and traceable throughout. The output goes beyond text — research reports, presentation decks, and runnable web apps published straight to hosting.
- Autonomous, fully transparent execution — the task plan checks itself off in real time, and every tool call and observation can be replayed
- Plugin ecosystem — web search, code sandbox, private datasets, Gmail and more — connectors and skills, freely combined
- Deliverables hub — publish generated apps in one click, with dashboards, databases, and file storage built in
- Scheduled tasks & cloud computer — headless runs on a schedule, plus a persistent workspace ready whenever you are
Your private data, searched by meaning
Upload JSONL or CSV and the index builds itself: BM25 keyword and semantic vector recall fused by RRF ranking, LLM answers with citations, and multimodal image-text search out of the box.
- Hybrid recall — keyword + semantic vectors fused with RRF, plus query completion and web-augmented search
- Cited AI answers — streams an answer over the hits, with citations you can trace back to the source
- Multimodal — image-text embeddings and reranking — search images with text, and text with images
A different feed for every user
Give it a user profile, get a personalized feed back: interest-based, trending, and cold-start recall fused together, an LLM-written reason on every pick, and related items to cover "people also viewed" scenarios.
- Multi-channel recall fusion — interest / trending / cold-start quotas with tunable ratios and diversity shuffling
- Explainable picks — an LLM writes a one-line "why this was picked for you" on every recommendation
- Related items — similar-item picks on detail pages, even for brand-new items not yet indexed
Fine-tune a model of your own
From sample management to SFT post-training to deployment, the whole pipeline runs on the platform. Deploy the finished model as an OpenAI-compatible endpoint in one click — ready for search, recommendations, and agents to call.
- Samples to training, end to end — fine-tuning sample management, job orchestration, and GPU status at a glance
- OpenAI-compatible deployment — /v1/chat/completions · embeddings · rerank, all on the standard protocol
- Private models — connect your DeepSeek or Claude credentials, encrypted at rest
Built for your business scenarios
One engine, ready out of the box for your industry — combine search, recommendations, and agents however you like.
Chat your way to a production website, published and hosted in one click
Learn moreFrom outline to finished deck, exported as real PowerPoint (.pptx)
Learn moreStyle presets for posters, landing pages, and infographics
Learn moreStoryboard to animated HTML video, exported in one click
Learn moreProduct search and 1:1 recommendations in one stack, lifting conversion and basket size
Learn morePersonalized feeds and related picks that surface great content
Learn moreSemantic search over private docs, with agents that research and write
Learn moreOpenAI-compatible APIs in one line, with hosting for your own models
Learn moreOne API Key
for every capability
Search, recommendations, agents, and the inference engine are all exposed over REST / SSE, with chat and embeddings on the standard OpenAI protocol — point your base_url at api.agentsdance.ai/v1 and you're live.
curl https://api.agentsdance.ai/v1/chat/completions \
-H "Authorization: Bearer $AGENTSDANCE_API_KEY" \
-d '{
"model": "agentsdance-llm-sft",
"messages": [
{"role": "user", "content": "Explain vector search"}
],
"stream": true
}'Ready to hand things off to AI?
The free tier covers the full agent, search, and recommendation experience — no credit card required.