Text → Aexol (From Tasks)

From Tasks are the reverse of code generation: instead of generating code from an Aexol spec, you describe your application in natural language and the AI writes the .aexol specification for you. Think of it as "spec-first design, AI-assisted."

How It Works

Plain English Description
         │
         ▼
   AI Generation (Claude/GPT)
         │
         ▼
   Parser Validation
         │
    ┌────┴────┐
    │  PASS?   │
    └────┬────┘
    No   │   Yes
    ▼    │    ▼
  Retry  │  .aexol Spec
  (up to │  Ready for
  max    │  Code Gen
  retries)│

The AI iteratively refines the output — if the generated Aexol doesn't parse or validate, it retries with the validation errors as feedback until the spec is syntactically and semantically correct.

Starting a From Task

Via the agent or MCP API

Start a from task through the backend's MCP API — the agent app (with the Studio Binding extension) calls the same endpoints. Enter:

Via Remote MCP

curl -X POST "https://api.aexol.ai/mcp" \
  -H "Authorization: Bearer sk-aexol-team-..." \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "remote_start_from",
      "arguments": {
  "content": "A task management app with projects, tasks, and users. Tasks have status: todo, in_progress, done. Users can be assigned to tasks.",
  "guidance": "Use camelCase for field names",
  "maxRetries": 3
}
    }
  }'

Check status:

curl -X POST "https://api.aexol.ai/mcp" \
  -H "Authorization: Bearer sk-aexol-team-..." \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "remote_get_from_task",
      "arguments": { "taskId": "task_..." }
    }
  }'

Wait for completion:

curl -X POST "https://api.aexol.ai/mcp" \
  -H "Authorization: Bearer sk-aexol-team-..." \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "remote_wait_task",
      "arguments": { "taskId": "task_...", "taskType": "from" }
    }
  }'

Example

Input (natural language):

A blog platform with authors and posts.
Authors have a name and bio.
Posts have a title, content, and publish date.
Each post belongs to one author.
Posts can be in draft or published state.

Output (.aexol):

type Author {
  id: string
  name: string
  bio: string
  posts: [Post]
}

type Post {
  id: string
  title: string
  content: string
  publishDate: date
  author: Author
  status: PostStatus
}

enum PostStatus {
  draft
  published
}

Refinement

Already have an Aexol spec but want to improve it? Use refinement instead of generating from scratch:

curl -X POST "https://api.aexol.ai/mcp" \
  -H "Authorization: Bearer sk-aexol-team-..." \
  -d '{
    "jsonrpc": "2.0", "id": 1, "method": "tools/call",
    "params": {
      "name": "remote_refine_aexol_content",
      "arguments": {
  "aexolContent": "type User { id: string, name: string }",
  "refinementPrompt": "Add email field and role enum with admin/editor/viewer values"
}
    }
  }'

You can also refine an existing From task — useful when the first generation was close but needs adjustments.

Knowledge Base Integration

When your project has knowledge-base documents uploaded via the backend API, the AI automatically retrieves relevant chunks and includes them in the generation prompt. This means generated specs reflect your team's existing types, naming conventions, and domain model.

From Task Lifecycle

Progress can be polled in real time via the MCP API (remote_wait_task / remote_get_from_task) — the response includes the current attempt count and validation errors as they are retried.

Next Steps

• Aexol CLI — Generate code from your new Aexol spec
• Agent Overview — Run spec work end-to-end in the agent app
• Billing & Credits — How usage is metered