Quickstart

Use Tentarc to run your first “real execution” task, not just chat.

This page follows the app path: install → model connection → workspace → services → first execution.

Note Tentarc is a local-first execution platform. From day one, split personal, team, and production workflows into separate workspaces.

Setup

1. Install and open Tentarc

After installation, open the app.

2. Connect Agent API

Open Settings → Agent API and choose one connection method first (preset providers below):

• Claude Subscription: sign in with subscription (OAuth)
• Anthropic: direct Anthropic API key
• OpenRouter: multi-model aggregation (single key)
• Vercel AI Gateway: gateway access (single key)
• MiniMax
• MiniMax (EN)
• Zhipu GLM
• Z AI
• Kimi For Coding
• Volcengine
• Ollama: local models
• OpenAI
• Groq
• Together
• SiliconFlow

No API connection, no execution capability.

3. Create or select a workspace

One boundary, one workspace (personal / work / production).
Each workspace isolates:

• services (MCP / API / local folders)
• permission modes
• skills
• sessions, memory, and automations

If you used Tentarc before, you can select an existing workspace; for a new project, creating a new workspace is recommended.

4. Connect services and one task ingress

Open Services and connect at least one data source:

• Local folder (recommended first): connect a git repository or docs directory
• MCP server: GitHub / Linear / Slack / Notion / self-hosted
• API: Gmail / Calendar / custom OpenAPI

If you want to dispatch tasks from anywhere, open Apps and connect one channel ingress: Telegram / Discord / Lark / Slack.

5. Start with one-click Quick Prompts

After the basic setup, don’t rush into writing long prompts. Run one Quick Prompt first to feel the reusable workflow.

You’ll see several Quick Prompt cards on the home page (for example: bug scan / PR description / progress board):

1. Click one card: the prompt is auto-filled into the composer.
2. Pick the right context: confirm in the composer footer
   • Work in Folder: points to your project directory (decides which repo to read/write)
   • Choose Services: selects external services (GitHub/Notion/Slack…; local-repo tasks still work without services)

6. Run your first agent task

Press Enter and execute directly.

1. Adjust 1–2 lines if needed: complete repo scope, output format, and constraints (e.g. “minimal patch only”).
2. Press Enter to send: run the first pass first; don’t open another session at the same time.
3. Use turn-queue to control pacing while running:
   • Keep queueing on by default (prefer followup) so appended requests don’t interrupt current execution.
   • To steer without interruption, use steer (Cmd/Ctrl + Enter).
   • If current execution is clearly off track, use interrupt (Cmd/Ctrl + Shift + Enter) to take over immediately.

7. Set permission mode

Recommended starting modes:

• Explore: read-only
• Ask to Edit: confirm before mutation

Use Auto only for trusted, repetitive, low-risk workflows.

In the composer, use Shift + Tab to switch mode.

8. Reuse results with Skills and Automations

If a flow repeats twice, it should be productized:

• add skill-creator in Skills
• call it with @ in chat
• describe the workflow you want
• convert stable flows into Automations (scheduled or trigger-based)

After the first successful run, choose one recurring flow and turn it into an Automation.