Use Kimi more confidently with task-focused guides built from official documentation, clearly dated observations and reproducible methods. Start with the page that matches what you are trying to do—not the Kimi feature with the loudest name.
Independent publication: Kimi AI Guide is not affiliated with, endorsed by or sponsored by Moonshot AI. Official links are identified as such; our analysis is labeled separately.
Start with Kimi
How to Use Kimi AI Chat
The practical starting point for asking questions, researching, working with sources and choosing between the current model options. It includes a reusable prompt brief, an answer-verification checklist and a guide to starting clean sessions.
Best for: first-time users, researchers, writers and anyone comparing Chat with Agent tasks.
Kimi AI Prompts: Templates and Browser Prompt Builder
Build a structured prompt locally in your browser, copy it into Kimi and use focused templates for research, files, writing, coding and data work. The builder sends no request to Kimi and stores no entered text.
Best for: users who want a practical prompt starting point without handing their draft to another service.
Kimi AI Login: Official Access & Account Help
Use the correct Kimi website, distinguish the consumer app from the developer console and troubleshoot common sign-in problems without exposing credentials or verification codes.
Best for: new accounts, returning users and people who are unsure whether a Kimi link is official.
Kimi AI App: Official Downloads and Safety Check
Find the verified Kimi routes for web, iPhone, iPad and Android, check the store publisher and identifier, and keep the mobile app separate from the Kimi Work desktop Agent. Unresolved HarmonyOS listing details remain explicitly unclaimed.
Best for: readers who need the official app or want to screen a store listing before installation.
Research, files and multi-step work
Using Kimi AI with Files and Long Documents
Plan a document task before uploading, create a coverage ledger for long files, request traceable citations and verify summaries against the source. The guide also explains Kimi’s currently documented file limits and privacy considerations.
Best for: PDFs, reports, contracts, research papers, spreadsheets and multi-file comparisons.
Kimi Search: Sources, Image Search and Verification
Use ordinary Kimi Search for a bounded current-information task, inspect source identity and support, and apply a reusable verification worksheet. Measured citation results remain on the separate Search Citation Accuracy Test rather than being duplicated here.
Best for: current web questions, direct-URL analysis, image-origin searches and readers checking whether a cited page supports the answer.
Kimi Vision: Images, OCR, Charts and Video Test
Inspect one controlled run across printed text, handwriting-style text, a table, chart, interface, scene and short video. The frozen test passed 35 of 36 atomic checks after the completed response was routed to K2.6 Instant; it is not a K3 or product-wide score.
Best for: OCR, chart and interface reading, visual extraction and short-video understanding with explicit evidence limits.
Kimi Agent: Capabilities, Limits and Access Check
Understand when Agent fits a multi-step task, how it differs from ordinary Chat and Agent Swarm, and which documented context, output and credit limits apply. The page reports its dated access blocker separately from performance.
Best for: tool-based tasks, deliverables and users deciding between Chat and Agent.
Kimi Agent Swarm: How It Works and Test Status
Separate Swarm’s documented parallel-agent workflow from Standard Agent and inspect the preregistered comparison protocol. The signed-in entry point was observed, but no task was submitted and no performance score is claimed.
Best for: large, divisible research batches and readers checking whether Swarm access and credits are actually available.
Kimi Projects: Files, Instructions, Memory and Context
Organize reusable reference files, persistent instructions and separate chats inside one Project. The guide includes a disclosed two-chat test using an original synthetic fixture.
Best for: recurring research, reusable project rules and work that depends on the same source files across chats.
Kimi Memory: What Persists Across Chats?
Distinguish selected saved information from current-chat context and Projects, then inspect our completed synthetic recall, update, deletion and false-memory test. All five individual fresh-chat recalls were exact, the update and visible deletion checks passed, five never-saved fields returned NOT STORED, and targeted cleanup completed for a total of 100/100 on August 6, 2026.
Best for: users who want information reused across chats and need a clear way to verify, update or remove it.
Kimi Deep Research: Access Test and Limits
Learn how Kimi documents Deep Research, how to request a source-auditable report and what our dated access attempt could—and could not—establish.
Best for: extended web research, source verification and users comparing Chat search with a longer research workflow.
Kimi Context & File Fit Checker
Estimate whether pasted text or a locally selected plain-text file may fit a documented Kimi Chat, API or Code context before upload. The result is a planning estimate, not an exact tokenizer or parsing guarantee.
Best for: long prompts, large text files and deciding whether to split material.
Create and edit deliverables
Kimi Websites: Builder, Export and Access Test
Plan a site with observable requirements, inspect the generated preview, separate revisions from publication and verify exported code before handoff. The page preserves its dated high-demand blocker as an access result, not a failed quality test.
Best for: landing pages, prototypes and users who need a testable website brief without publishing the preview automatically.
Kimi Docs: Word Review, Revisions and PDF Test
Review an existing Word document while checking corrections, comments, revision suggestions, tables, links and DOCX/PDF export separately. The original 16-error fixture and scoring method are available even though the review run was blocked before generation.
Best for: Word review, document revisions, structured comments and export-integrity checks.
Kimi Sheets: Formulas, Charts and Accuracy Test
Use a controlled CSV and known answers to evaluate formulas, totals, missing values, duplicate handling, charts and XLSX integrity. The current report states that workbook generation did not begin during the recorded attempt.
Best for: spreadsheet analysis, formula verification, chart creation and downloadable-workbook QA.
Kimi Slides: K3 Workflow, Exports and Test Status
Prepare a source-controlled slide brief, use the current K3 workflow and inspect fact preservation, layout, editability and PPTX export as separate checks. The public fixture and scoring rubric remain distinct from the blocked run status.
Best for: editable presentations, source-backed decks and PowerPoint export review.
Work locally and reuse procedures
Kimi Work: Desktop Agent, Permissions and Safe-Folder Test
Set up the desktop Agent against a disposable folder, choose a permission mode, preserve before-and-after hashes and inspect local file safety independently. The recorded turn stopped before a first token or tool action, so the page does not claim that permission enforcement passed.
Best for: Windows or macOS desktop workflows, local folders, scheduled tasks and permission-sensitive work.
Kimi Claw: Setup, Memory, Files and Automation
Understand Claw deployment, persistent Memory, Workspace files, Skills, terminal access and HEARTBEAT schedules. Controls were visible, but no usable authorized instance was confirmed and nothing was deployed, linked or changed.
Best for: readers evaluating a persistent hosted assistant and the account, deployment and safety boundaries around it.
Kimi WebBridge: Browser Setup, Permissions and Safe Test
Learn how the local bridge and browser extension interact with existing Chrome or Edge sessions, then inspect a twelve-check localhost protocol. WebBridge was not installed or connected, so fixture values remain ground truth rather than results.
Best for: browser automation users who need a clean-profile test, explicit action boundaries and existing-session risk controls.
Kimi Scheduled Tasks: Timing, Notifications and Missed Runs
Separate cloud Kimi scheduling from local Kimi Work behavior, then inspect our dated one-task cloud pilot. Creation, editing, pause, resume and cleanup persisted. Run once now returned a service-busy error, and the authoritative scheduled attempt advanced to tomorrow without exposing a new history row, result conversation or notification inside the frozen five-minute window.
Best for: recurring prompts and desktop workflows that need auditable timing, notification and cleanup behavior.
Kimi Presets: Setup, Limits and Consistency Test
Learn how Kimi documents reusable Preset text and inspect a frozen five-pair comparison protocol. The signed-in account exposed no Presets entry point, so no Preset was created and no message or comparison ran.
Best for: users evaluating saved prompt text and the boundary between Presets, Projects, Memory, Skills and scheduled tasks.
Kimi Skills: Creation, Usage and Consistency Test
Turn a reusable procedure into a candidate Skill, distinguish an exported file from an installed Custom Skill and compare repeated tasks against the same frozen rubric. The five-task comparison remains not run because the real Skill creator was blocked.
Best for: reusable SOPs, repeatable Agent behavior and readers deciding whether a Skill adds value beyond an ordinary prompt.
Kimi Plugins: Permissions, Credits and Safe Test
Map plugin availability, OAuth and permission boundaries, then inspect a low-risk public-data protocol. The World Bank plugin was selectable, but a subscription-capacity gate stopped the attempt before chat creation or plugin invocation, so no accuracy or cost result is claimed.
Best for: readers deciding whether an external service, permission request or credit charge fits a safe Kimi workflow.
The three-step learning path
If you are completely new to Kimi, use the guides in this order:
- Access: Confirm that you are using an official Kimi service and understand which account context you need in the login guide.
- Control: Learn how model choice, task definition, output format and source checking affect results in the chat guide.
- Ground: When your answer must depend on specific evidence, use the structured upload and verification workflow in the files guide.
Quick decision table
| Your situation | Recommended guide | First action |
|---|---|---|
| You want a reliable first prompt | Chat | Write the task, audience, scope and desired output |
| You want to assemble and copy a structured prompt locally | Prompt Builder | Choose a template, review every field and remove private information before copying |
| A search result shows several Kimi domains | Login | Navigate directly to the verified official domain |
| You need the official web or mobile app | Kimi AI App | Check the store identifier and publisher before installing |
| Your PDF is long or contains scans | Files | Inspect the file and define a coverage plan before upload |
| You need current facts from the web | Search | Require links, dates and a source-to-claim table, then open the cited pages |
| You need text or values extracted from an image, chart, interface or short video | Vision | Define exact fields and independently check every returned value |
| You need to compare multiple documents | Files | Name each file and define the comparison fields |
| You need a generated website preview and export | Websites | Freeze acceptance conditions and keep publishing as a separate decision |
| You want Kimi to review a Word file | Docs | Preserve the original and define corrections, comments and export checks separately |
| You need formulas, totals or charts in a workbook | Sheets | Provide known values and inspect formulas rather than displayed numbers alone |
| You need an editable presentation | Slides | Freeze the source facts, slide structure and PPTX acceptance checks |
| You need controlled access to local files | Work | Use a disposable folder, the narrowest permissions and before-and-after hashes |
| You want to reuse the same operating procedure | Skills | Compare the Skill with the same instructions used as an ordinary prompt |
| You need an API key or code example | Developers | Use the developer hub; consumer-login steps are not enough |
| You are choosing a Kimi model | Models | Match the model to the product surface and task |
| You have a large task that can be split into parallel research streams | Agent Swarm | Freeze the source list, output schema, access state and credit ceiling before submission |
| You want selected information reused across chats | Memory | Define what should persist and test recall, correction and deletion separately |
| You want to insert the same editable prompt text repeatedly | Presets | Confirm the Presets control exists before planning a comparison or migration |
| You need a persistent hosted assistant with files and schedules | Claw | Confirm an authorized instance and deployment boundary before changing anything |
| You want an Agent to act inside a browser session | WebBridge | Start with a clean profile and a localhost-only fixture |
| You want Kimi to call an external service | Plugins | Review permissions, OAuth, credits and side effects before the first invocation |
| You need a prompt to run later or repeatedly | Scheduled Tasks | Record timezone, notifications, edit, pause and missed-run behavior |
Build a better Kimi prompt
Kimi’s official prompt guidance recommends clear context, a specific task, an output format and constraints. Our standalone Kimi AI Prompt Builder turns those elements into an editable prompt in your browser and includes a small set of task-focused templates.
The tool is deterministic, requires no Kimi API and does not save your entries. Review the assembled text before copying it, remove confidential material and independently verify high-impact answers. The full templates and builder live on that page rather than being duplicated on this hub.
The Kimi answer-check card
Use this card after any research, numerical or document-based response:
- [ ] The answer addresses the exact question.
- [ ] Each important factual statement has an inspectable source.
- [ ] The cited page actually supports the statement.
- [ ] Dates, names, versions and units match the source.
- [ ] Calculations were recomputed independently.
- [ ] Missing evidence and uncertainty are visible.
- [ ] High-stakes conclusions were reviewed by a qualified human.
Fluency is not verification. Kimi’s Terms of Service explicitly caution users to assess output accuracy and appropriateness independently.
What our guides mean by “verified”
We use narrow evidence labels so that documentation and testing are not blended together:
| Status | What we did | What it does not prove |
|---|---|---|
| Documentation checked | Read and linked a current official Kimi or Moonshot AI source | That the feature works identically for every account or region |
| Access observed | Confirmed that a dated account or interface could reach a feature | Output quality or reliability |
| Workflow tested | Ran a disclosed input through a defined process and recorded the result | Performance outside that input and setup |
| Not tested | No suitable login, plan, budget or safe test material was available | Nothing beyond the cited documentation |
Every test should publish its date, account tier, model, input, steps, scoring rule and limitations. When those details are absent, treat the statement as a guide—not a benchmark.
Browse the Kimi AI Test Lab for dated result records, original evidence and explicit test limitations.
Current product map
The official Kimi Help Center currently describes a consumer Chat experience, Agent tools, Kimi Code, Kimi Work and a separate Kimi Open Platform for developers. These are related, but instructions and billing do not always transfer from one product to another.
- For consumer conversations, start at the official Kimi website.
- For official product instructions, use the Kimi Help Center.
- For developer integration, continue to our developer hub and the official Open Platform.
- For plan comparisons, use our pricing hub rather than assuming Chat and API charges are interchangeable.
Privacy before productivity
Before uploading a file or pasting confidential material:
- remove personal, financial, health, legal and authentication data that the task does not require;
- check who owns the material and whether you are allowed to upload it;
- read the current Kimi data-usage terms for the product you are using;
- avoid third-party plugins unless you understand their permissions and policies; and
- use synthetic or redacted material when testing a workflow.
Kimi’s Help Center says consumer conversation data may be used for model training after described safeguards and provides an account-level opt-out route through official support. It separately states that enterprise and API data have different training treatment. Always check the current policy rather than assuming that one Kimi product’s rule applies to another.
Continue through the reference
This hub focuses on practical use. Continue to the section that matches your next decision:
- compare current model roles in the Kimi models hub;
- separate membership, API and Business costs in the pricing hub;
- install the CLI or build with the API from the developer hub;
- inspect dated runs and raw evidence in the Kimi AI Test Lab; and
- report an outdated fact through Sources & Corrections.
The goal is a connected reference, not a pile of overlapping articles.
Frequently asked questions
Which Kimi AI guide should I read first?
Start with the official login guide if you have not used Kimi before. If you are already signed in, use the chat guide. Use the files guide whenever the answer must be grounded in a document you provide.
Are these official Kimi guides?
No. They are independent guides from KI AI Team. We link to official Kimi documentation and clearly distinguish it from our own analysis.
Do the guides include live tests?
Only when a page explicitly labels a workflow as tested and provides the date, setup, input, method and limitations. Otherwise, the page is documentation-based.
How can I report an error?
Email [[email protected]](mailto:[email protected]) with the page URL, the statement you believe is wrong and a reliable supporting source. Accepted corrections should be reflected in the page’s update record.
Official sources for fact-checking
All sources accessed August 4, 2026.