Product Guides

Choose models and connect your apps.

Use Models to see the aliases currently available to your account. Use Connectors to authorize the services Kendr may read or act on for your tasks.

Open Models Open Connectors

Choose a Kendr model

The Models screen shows each public alias, capability, context size, current credit rate, and availability. Choose the alias by outcome rather than assuming a vendor model will always remain available.

PathAliasUse it forWhat to expect
Exact modelkc-…Direct chat, generation, structured output, vision, web search, and compatible tools.A stable Kendr alias maps the request to an available model route.
Kendr Intelligentkendr-intelligentGeneral-purpose managed routing with optimized context.Kendr selects an eligible general model and applies Kendr Optimizer.
Kendr Intelligent Directkendr-intelligent-directGeneral-purpose managed routing with original context.Kendr selects from the same general route category without optimizing the request context.
Kendr Coderkendr-coderCoding work through any compatible harness, with optimized context.Kendr restricts selection to coding-focused models and applies Kendr Optimizer.
Kendr Coder Directkendr-coder-directCoding work through any compatible harness, with original context.Kendr restricts selection to coding-focused models without optimizing the request context.
SpecializedFeature-specificCapabilities such as media, search, or other purpose-specific workflows.The request and response contract depends on that feature.

All four Kendr managed routes accept images and rendered scanned-PDF pages when they are listed as available. Availability requires an executable physical route with an explicit OCR attestation; Kendr Coder additionally requires coding, OCR, and image input on that same physical route.

curl https://api.kendr.org/v1/models \
  -H "Authorization: Bearer $KENDR_API_KEY"
Runtime discovery

Read GET /v1/models when your application starts or refreshes its model picker. Do not assume every alias is available to every account at all times.

Web search capability

Text aliases can expose web_search. Use web_search: false on a model request when a prompt should not use current web information.

Supported model-provider families

Kendr aliases may be backed by the provider families below. Your application still sends a Kendr key and Kendr alias; it does not send a separate provider credential.

Provider keyFamily
bedrockAmazon Bedrock-compatible model routes
openaiOpenAI model families
anthropicAnthropic Claude model families
googleGoogle Gemini model families
xaixAI model families
deepseekDeepSeek model families
zaiZ.ai model families
qwenQwenCloud Qwen models and hosted partner-model routes
byteplusBytePlus ModelArk Seed, DeepSeek, GLM, GPT-OSS, image, video, embedding, and 3D catalog identities
kimiKimi model families
mistralMistral model families
sarvamSarvam 105B native Indian-language models
tokenraTokenRa hosted chat routes, including OxAlpha, plus ArtSDance 2.5 behind the provider-neutral kendr-video product
TokenRa credential

An administrator can add or rotate the TokenRa key in the model-provider Connectors panel. Kendr stores that key encrypted; TOKENRA_API_KEY is only an optional deployment bootstrap fallback. Connect and discover before enabling an entitled, priced route.

BytePlus ModelArk credential

Add a ModelArk API key in the model-provider Connectors panel, then connect to discover the models activated for the account in ap-southeast-1. BYTEPLUS_API_KEY is the preferred optional deployment fallback and ARK_API_KEY is also accepted. Chat-capable identities use ModelArk's official OpenAI-compatible endpoint; non-chat identities stay task-classified and cannot be selected as chat routes.

One alias, multiple providers

When the same logical model is available through QwenCloud, TokenRa, Bedrock, Ollama, or a direct provider, Kendr keeps one public alias. It selects a healthy route using cost, reliability, and latency signals, then reuses that provider route for compatible requests. A failed or incompatible route triggers failover, and the successful replacement becomes the new affinity.

For Bedrock-backed model calls, BEDROCK_MODEL_REGION controls the model invocation region. Web search is provided through Kendr's managed web-search gateway, which is configured separately on the Kendr side and requires no setup in your application.

Connect an app to Kendr

  1. 1
    Open Connectors
    Sign in and go to Account → Connectors.
  2. 2
    Choose a service
    Select a service marked ready for connection and review what it can provide.
  3. 3
    Continue to the provider
    Sign in to that service and review the requested scopes on its consent page.
  4. 4
    Approve only what you need
    Complete consent only if the account, scopes, and Kendr connection are correct.
  5. 5
    Return to Kendr
    Confirm the connector now shows Connected, then run a read-only task first.
Actions remain explicit

A connected app can provide context and tools, but you should still review sends, edits, creates, deletes, purchases, and other consequential actions before allowing them.

App connector catalog

Google Workspace
Gmail, Drive, Docs, Sheets, Slides, Contacts, Tasks, and Calendar through one administrator-configured Google OAuth application.
Microsoft 365
Outlook Mail, Outlook Calendar, OneDrive, SharePoint, Word, Excel, PowerPoint, and Teams through one Microsoft Entra application.
GitHub
Repositories, issues, pull requests, Actions, and development context.
Slack
Workspace search, channels, threads, and approved message actions.
Notion
Read-only shared-page search and retrieval through Kendr's adapter, limited to workspace content granted to the connection.
Linear
Read-only issue, project, and delivery context through the connected Linear workspace.
Todoist
Read-only tasks, projects, sections, comments, labels, reminders, and account details using the data:read scope.
Atlassian Cloud
Jira and Confluence context for projects, work items, and documentation.
Salesforce
Customer and CRM context within the scopes approved by the account.
Figma
Design files, comments, and product-design context when supported.
Airtable
Bases and structured records available to the connected account.
Daily-work connectors stay read-only

Google Contacts, Google Tasks, Microsoft Word, Excel, PowerPoint, Linear, Notion, and Todoist expose search and retrieval tools only. Google Tasks, Notion, and Todoist are served through Kendr's read-only adapters; the remaining services use provider read scopes or read-only tool endpoints.

Google's People API MCP server is a Developer Preview. Workspace-directory search depends on the connected Google Workspace tenant and preview availability.

GitHub OAuth scope

Read-only chat tools describe what Kendr exposes to the model. GitHub currently uses a broad repository OAuth grant even though Kendr connects it to GitHub's read-only MCP endpoint. Administrators should review the exact displayed scope before enabling GitHub.

Disconnect or reconnect

  • Open Connectors and choose Disconnect when Kendr should no longer use that service.
  • Revoke Kendr from the service's own connected-app settings when you want to end the authorization at the provider as well.
  • Reconnect after a token expires, permissions change, or you need to use a different account.
  • If a task cannot find expected data, confirm the connected identity and granted scopes before adding broader access.