Marketing automation. Autonomous research. An agent control plane. Plus meetings, classes, tasks, English, stocks and games — one tool per scenario. What ties them together is the quiet infrastructure: observability, token economics, skill evolution, shared memory.
A knowledge graph that grows as you ask
12-class tagging · doc-to-graph
Long-form, in both languages
348 bilingual · 32 series
A research agent that never sleeps
66K+ papers · 373 experiments
Docs distilled into Agent Skills
12 products · 1,000+ skills
Agents dispatched across machines
4 machines · 15 products
Idea to shippable game in 30 min
code + art + cutscenes, auto
One sentence, one campaign
172K LOC · 11 cloud services
One sentence in, a full campaign out: trends, copy, imagery, short video, voice-over and 9-language localization run end to end. Spark Drama Studio handles agentic screenwriting, directing and review; GEO optimization gets the result cited by AI search.
Distill a company's docs into agent-ready Skills.
Point it at a folder of product docs; an LLM writes the detail and intent Skills an agent needs, ready for Claude Code, Cursor or any SKILL.md reader. Zero-dependency, with a GEO feedback loop and auto-generated MCP servers.
Pick a template and a style, hit generate: code, art and cutscene video are produced end to end. The elevator runtime plans, executes and verifies, with 7 guardrail categories and visual self-debug across 6 providers.
Official web games and AI-forged creations, all in one arcade. Play in the browser, add friends, compare scores and send challenges; anything made in MiniGameForge publishes here in one click.
Two parallel libraries, never translations: Chinese compresses, English expands. 32 series cover ML, mathematics, cloud and agent systems, with LaTeX, full-text search and a custom dark theme.
An autonomous research system that never stops.
One pipeline discovers and reads papers, grows a 107K-node knowledge graph and debates new ideas; another designs experiments, writes manuscripts and reviews them. Self-healing supervisors and explicit state machines keep every step auditable.
Dispatch long-running agents across machines.
Goals become a DAG of verifiable subgoals, with planning, execution, review and verification split across different models. Agents run in bwrap/Landlock sandboxes on their own machines; failing routes cool down, budgets and approvals are explicit, and a watchdog requeues stuck jobs.
Ask a question. Watch a knowledge graph grow.
Every answer spawns related concepts around it, and any sentence becomes a context-linked follow-up. Upload documents to build a private map the model can use as scaffolding, then export it as Markdown or JSON.
While the meeting runs, a quiet sidebar listens, records, and answers. System audio in, streaming transcript, rolling minutes and suggested replies grounded in your knowledge base, the web and your own APIs. Passive by design.
You attend class, AI takes the notes. Lecture audio is transcribed live, then organised into illustrated handouts and a topic-by-topic review path, so exam prep reads the handout instead of re-watching the video.
Tasks, schedule, notes and focus on one page. Todo lists break down to today, focus timers record real time invested, and courses, notes and deadlines share one timeline across phone and desktop.
It remembers the collocations and mistakes you miss. Every session leaves evidence, and tomorrow's lesson grows out of yesterday's errors; vocabulary, collocations and grammar each get their own track while a learner model adjusts difficulty.
A desktop cat named Clawd that watches every agent on your machine. Running, waiting, stuck or erroring — expressions and speech bubbles say what is happening. Seven skins, zero dependencies, one command.
Live A-share quotes with AI commentary you can check. Moving averages, volume and pattern indicators compute in real time, and the model turns them into plain-language support, resistance and the signals to watch today.
LLM App Security A practitioner's guide to securing web applications in the AI coding era: prompt injection, privilege escalation, data exfiltration and supply-chain poisoning, with defenses drawn from production incidents.
Read the notesNo projects match this filter.
How should a long-running agent split its budget across providers? Which steps are worth the most expensive model?
Which failures can compress into reusable skills, and which can only be discarded?
When multiple agents share one memory, how strict should the type system be?
From demo to production — the last mile: observability, replay, governance, and the mechanics of user trust.