350 / 350
Start from the problem you actually have. Pick a scenario, then narrow it down — every card says what it solves and when to reach for it.
119
Writing, meetings and documents. Pick the moment you need help in, rather than the product that has the most features.
Work that dies when you close the tab. A dot runs on its own cloud computer across ChatGPT, Slack, and Teams, and does read-only research while you're away.
Work that runs for hours and does not need you watching the chat.
More
- Vendor
- OpenAI
- Released
- 2026-09-29
- Capabilities
- in-product · chatgpt
- Where
- Inside another product
An errand is mostly logging in and clicking. Muse uses your existing accounts to do that part.
Shopping and email you would otherwise click through yourself.
More
- Vendor
- Meta
- Released
- 2026-09-08
- Capabilities
- personal-agent · proactive · shopping · email
- Where
- Inside another product
A job that crosses roles has to be held together by whoever assigned it. 万有无界 hands the same job to several digital employees in turn.
A cross-role assignment you would rather hand to a chain of agents.
More
- Vendor
- Wanyo (万有)
- Released
- 2026-09-02
- Capabilities
- multi-agent · enterprise · a2a · workbench
- Where
- Standalone app
Analyzing a folder of spreadsheets means uploading it to a web page and downloading the result back. Raccoon reads the files where they sit.
When the material cannot go up to a public site, and you still want it summarized.
More
- Vendor
- 商汤
- Released
- 2026-08-25
- Capabilities
- desktop-agent · data-analysis · knowledge-base · private-deployment
- Where
- Standalone app
It reads your Feishu chats and minutes, so it knows the real business context.
Teams already living in Feishu who want an agent.
It plans and executes across apps on your phone, not just answers.
If your phone, car and home gear are all Xiaomi.
104
No-code builders, workflow platforms, or code frameworks. The deciding factor is who maintains the thing after launch.
Work that dies when you close the tab. A dot runs on its own cloud computer across ChatGPT, Slack, and Teams, and does read-only research while you're away.
Work that runs for hours and does not need you watching the chat.
More
- Vendor
- OpenAI
- Released
- 2026-09-29
- Capabilities
- in-product · chatgpt
- Where
- Inside another product
RAG means running a vector database and an indexing pipeline. Compass parses documents, indexes them, and serves retrieval without either.
You need document retrieval for an agent and do not want to run the index yourself.
An errand is mostly logging in and clicking. Muse uses your existing accounts to do that part.
Shopping and email you would otherwise click through yourself.
More
- Vendor
- Meta
- Released
- 2026-09-08
- Capabilities
- personal-agent · proactive · shopping · email
- Where
- Inside another product
A job that crosses roles has to be held together by whoever assigned it. 万有无界 hands the same job to several digital employees in turn.
A cross-role assignment you would rather hand to a chain of agents.
More
- Vendor
- Wanyo (万有)
- Released
- 2026-09-02
- Capabilities
- multi-agent · enterprise · a2a · workbench
- Where
- Standalone app
Analyzing a folder of spreadsheets means uploading it to a web page and downloading the result back. Raccoon reads the files where they sit.
When the material cannot go up to a public site, and you still want it summarized.
More
- Vendor
- 商汤
- Released
- 2026-08-25
- Capabilities
- desktop-agent · data-analysis · knowledge-base · private-deployment
- Where
- Standalone app
It reads your Feishu chats and minutes, so it knows the real business context.
Teams already living in Feishu who want an agent.
79
Answer engines with citations versus retrieval you build yourself. Decide whether you need a product or a stack.
Work that dies when you close the tab. A dot runs on its own cloud computer across ChatGPT, Slack, and Teams, and does read-only research while you're away.
Work that runs for hours and does not need you watching the chat.
More
- Vendor
- OpenAI
- Released
- 2026-09-29
- Capabilities
- in-product · chatgpt
- Where
- Inside another product
RAG means running a vector database and an indexing pipeline. Compass parses documents, indexes them, and serves retrieval without either.
You need document retrieval for an agent and do not want to run the index yourself.
A job that crosses roles has to be held together by whoever assigned it. 万有无界 hands the same job to several digital employees in turn.
A cross-role assignment you would rather hand to a chain of agents.
More
- Vendor
- Wanyo (万有)
- Released
- 2026-09-02
- Capabilities
- multi-agent · enterprise · a2a · workbench
- Where
- Standalone app
Analyzing a folder of spreadsheets means uploading it to a web page and downloading the result back. Raccoon reads the files where they sit.
When the material cannot go up to a public site, and you still want it summarized.
More
- Vendor
- 商汤
- Released
- 2026-08-25
- Capabilities
- desktop-agent · data-analysis · knowledge-base · private-deployment
- Where
- Standalone app
A firm wants AI for its team and has nobody to run it. Agents execute in 360's isolated cloud sandbox; your code and data stay on your side.
You run a small business, want AI for your team, and won't stand up servers.
More
- Vendor
- 360
- Released
- 2026-07-28
- Capabilities
- enterprise-agent · multi-model · cloud-desktop · sandbox
- Where
- Standalone app
Drop in 50 sources and get a cited answer plus a finished deck.
Literature review, course prep, or a research brief.
69
For engineers who want an assistant inside the editor, in the terminal, or on a real issue. Start by where the agent is allowed to touch your machine.
Terminal only, and macOS. Inside that limit it reads and edits across files without a browser in the loop.
A coding agent you drive from the shell, without leaving it.
More
- Vendor
- Meta
- Released
- 2026-08-01
- Capabilities
- terminal · beta · multi-file · macos
- Where
- Standalone app
Every new conversation means re-explaining the project, and requirements still have to be typed. MiMo Code keeps the project in memory and takes spoken instructions.
One long-running project, and no interest in restating its context every session.
More
- Vendor
- 小米
- Released
- 2026-06-11
- Capabilities
- open-source · terminal · persistent-memory · voice-input
- Where
- Standalone app
Chat, research and code living in separate assistants means every switch drops your context. Vibe is one agent that does all three.
You want one assistant that handles office chat and also writes code.
More
- Vendor
- Mistral AI
- Released
- 2026-05-28
- Capabilities
- work-mode · code-mode · european · rename
- Where
- Standalone app
One agent working serially is the bottleneck on a large task. The work splits, and the agents run in parallel.
When the task is big enough that starting many agents beats watching one grind.
More
- Vendor
- xAI
- Released
- 2026-05-25
- Capabilities
- terminal · open-source · parallel · workflows
- Where
- Standalone app
General coding models handle HarmonyOS badly. CodeArts trained a model on Huawei's own HarmonyOS code, and states it does not store your code.
Teams under a compliance rule that keeps code inside the network, or building for HarmonyOS.
More
- Vendor
- 华为
- Released
- 2026-02-26
- Capabilities
- harmonyos · ascend · spec-driven
- Where
- Standalone app
Open code, local memory, and IM remote control in one desktop agent.
When you want an auditable agent, not a black box.
44
Generating pixels versus editing the layout you already have. These are different jobs with different tools.
It plans shots, generates and edits the cut, so you stay in one tool.
Commercial video with many revisions and many steps.
A current art-generation option from a vendor with real shipping experience.
You already use Meitu tools and want the new art line too.
A new option from a vendor that has shipped imaging tools for two decades.
You follow Meitu's product line and want the current release rather than the old editor.
It is a surface inside Grok rather than a separate app. Images and video generate from the same conversation, and the same models sit behind an API.
When you are mid-conversation and want to see the picture rather than describe it.
More
- Vendor
- xAI
- Released
- 2026-01-28
- Capabilities
- image-generation · video-generation · creative
- Where
- Inside another product
Removes the gap between a rough idea and a finished design file, so you stop assembling brand assets by hand.
Reaching for it when you need a full brand or marketing set in minutes, not just one image.
A Figma file cannot be shipped. Make turns the design — or a sentence — into working code you can run and edit, keeping the design context attached.
You are a designer who needs a working prototype, or a dev who wants to start from a real design file.
38
Hosted model access for building products. Judge on price, latency, context and data terms — not on leaderboard position.
Every vendor ships its own request shape. These model families come through an OpenAI-compatible endpoint, so the adapter already written keeps working.
You already speak the OpenAI API and want Meta's models without new glue.
More
- Vendor
- Meta
- Released
- 2026-07-09
- Capabilities
- openai-compatible · spark · image-generation
- Where
- Standalone app
Enterprises adopting large models end up buying deployment, fine-tuning and application tooling separately, then stitching them together. They are collected in one platform here.
When a company wants to build its own LLM applications without buying three separate toolchains.
More
- Vendor
- 零一万物
- Released
- 2026-05-21
- Capabilities
- agent-platform · multi-agent · 01-ai
- Where
- Standalone app
Every team wires agents its own way, and nobody can say who was allowed to touch production. Building, permissions and run history end up in one place.
Past a handful of internal agents, when someone will eventually ask who changed what.
More
- Vendor
- Released
- 2026-04-22
- Capabilities
- enterprise · orchestration · governance · rebrand
- Where
- Standalone app
A general model knows nothing about your corpus or your rules, so you assemble the dataset and the eval yourself. Forge covers training through deployment.
You need a model that serves your organisation alone and can justify its answers.
More
- Vendor
- Mistral AI
- Released
- 2026-03-17
- Capabilities
- fine-tuning · evaluation · sovereign · enterprise
- Where
- Standalone app
An agent that misbehaves only tells you after it ships. Studio runs the flow while you build, shows the evaluation, and lets you re-run any step's inputs.
Tuning an agent flow where you need to replay the same step fifty times.
More
- Vendor
- Mistral AI
- Released
- 2025-10-01
- Capabilities
- workbench · workflows · evaluation · agents
- Where
- Standalone app
Running Llama yourself means buying GPUs, writing the serving code and keeping it patched. This hands you an OpenAI-compatible endpoint, SDKs and fine-tuning instead.
When you want Llama behind an OpenAI-compatible API without running the infrastructure.
More
- Vendor
- Meta
- Released
- 2025-04-29
- Capabilities
- api · llama · openai-compatible · fine-tuning
- Where
- Standalone app
32
From a one-line prompt to a finished cut. The first question is whether you are generating footage or editing what you already shot.
It plans shots, generates and edits the cut, so you stay in one tool.
Commercial video with many revisions and many steps.
Pairs a track with visuals without a separate editing tool.
Artists who need a video for a release and work alone.
It is a surface inside Grok rather than a separate app. Images and video generate from the same conversation, and the same models sit behind an API.
When you are mid-conversation and want to see the picture rather than describe it.
More
- Vendor
- xAI
- Released
- 2026-01-28
- Capabilities
- image-generation · video-generation · creative
- Where
- Inside another product
Characters and props stay consistent across up to 100 episodes.
Short-drama and motion-comic studios producing series.
Turns a plain-language brief into finished visuals instead of a stack of retouch steps.
A small creative team that needs finished assets on a deadline without a designer.
Keeps a short film's look consistent across shots without manual continuity work.
A short film where visual consistency across shots matters.
24
Knowing whether a model or agent actually got better. Trace what happened before you tune anything.
RAG means running a vector database and an indexing pipeline. Compass parses documents, indexes them, and serves retrieval without either.
You need document retrieval for an agent and do not want to run the index yourself.
Every team wires agents its own way, and nobody can say who was allowed to touch production. Building, permissions and run history end up in one place.
Past a handful of internal agents, when someone will eventually ask who changed what.
More
- Vendor
- Released
- 2026-04-22
- Capabilities
- enterprise · orchestration · governance · rebrand
- Where
- Standalone app
An agent that misbehaves only tells you after it ships. Studio runs the flow while you build, shows the evaluation, and lets you re-run any step's inputs.
Tuning an agent flow where you need to replay the same step fifty times.
More
- Vendor
- Mistral AI
- Released
- 2025-10-01
- Capabilities
- workbench · workflows · evaluation · agents
- Where
- Standalone app
An agent demo dies in production because nobody can see what it did. AgentArts pairs multi-agent runtime with end-to-end observability and evaluation, which is the part enterprises actually buy.
You are shipping agents to production and need runtime plus observability together.
A single GPU is fine until it is not. Dynamo splits prefill and decode across nodes, routes around redundant work, and caches KV to cheaper storage.
You are serving LLMs across a multi-GPU cluster and prefill/decode is your bottleneck.
The agent answers on Monday and has quietly drifted by Friday. Tracing, evaluation and guardrails are what tell you which part moved.
Agents already in production, where you need to know why one answer changed.
More
- Vendor
- NVIDIA
- Released
- 2025-04-01
- Capabilities
- agent-ops · guardrails · evaluation · self-hosted
- Where
- Inside another product
23
Running models on your own machine or your own server. The reason is usually privacy, latency, or cost per token.
It plans and executes across apps on your phone, not just answers.
If your phone, car and home gear are all Xiaomi.
You get an OpenClaw-style agent with system permissions, no key management.
HarmonyOS phone or tablet owners doing office tasks.
Agents are walled inside their own interface, and data lives on someone else's server. OpenClaw runs on your machine and answers on WhatsApp, Telegram, Discord, Slack, or Signal.
You want the agent on your own hardware, driven from a chat app you already use.
More
- Vendor
- OpenClaw
- Released
- 2026-01-15
- Capabilities
- self-hosted · gateway · messaging
- Where
- Standalone app
Some regulators will not let agent traffic leave the building. North runs in your own VPC, on-prem, or in Cohere's Model Vault.
A compliance regime that names the infrastructure your agents are allowed to run on.
More
- Vendor
- Cohere
- Released
- 2025-01-10
- Capabilities
- on-premise · sovereign · government · connectors
- Where
- Standalone app
Most coding agents stop at suggesting. Goose installs, executes, edits and tests — and says so itself: not just for code, but research, writing, automation and analysis.
When you want the same run from a terminal, a desktop app, or your own code.
More
- Vendor
- Block
- Released
- 2025-01-01 (month only)
- Capabilities
- open-source · linux-foundation · mcp-extensions · local-first
- Where
- Standalone app
La Plateforme is the API. AI Cloud is the compute underneath it — sovereign and self-hosted deployments for teams that cannot send data out.
You need Mistral models on infrastructure you control, for sovereignty reasons.
13
Speaking, singing, or turning speech into text. Voice cloning and dubbing are a different market from plain TTS.
Pairs a track with visuals without a separate editing tool.
Artists who need a video for a release and work alone.
Produces a usable backing track without a composer or a session musician.
A video that needs original background music cleared for use.
AI usually lives in a tab you have to go find. Apple Intelligence runs on-device and shows up inside Mail, Messages and notifications — plus a redesigned Siri.
You want AI in the phone's own apps, on-device, rather than in another app to switch to.
More
- Vendor
- Apple
- Released
- 2024-10-28
- Capabilities
- in-product · ios
- Where
- Inside another product
Synthesized speech reads like an announcement. Hume lets you direct the emotion it performs in, and clone a voice you own.
Audio where the voice has to sound like it is feeling something, not reading.
More
- Vendor
- Hume AI
- Released
- 2024-06-19
- Capabilities
- emotion-aware · voice-clone · tts · expression-control
- Where
- Standalone app
Licensing a track for a video or a game takes weeks and costs more than the project. The music is generated instead, cleared to ship.
When you need a music bed that fits the cut and will not arrive with a rights dispute.
More
- Vendor
- Uncharted Labs
- Released
- 2024-04-10
- Capabilities
- music-generation · audio · creative
- Where
- Standalone app
Voice agents sound broken because speech generation is too slow. It starts speaking in under 100 milliseconds.
When you are building a voice agent that must answer without an audible pause.
More
- Vendor
- Cartesia AI
- Released
- 2024-01-01 (month only)
- Capabilities
- tts · low-latency · voice-clone · on-device
- Where
- Standalone app