Aflare

Overview

aflare is an AI-powered terminal workflow engine that generates and executes multi-step automation pipelines from natural language descriptions. It supports a broad catalog of built-in agent nodes (run aflare list for the full set), including LLM providers, code analysis, file operations, web fetching, and system integrations. Workflows are defined as deterministic YAML files that can be version-controlled and reused.

Key features:

• Natural language to YAML workflow generation
• Built-in nodes covering LLM, fetch, execute, transform and more (aflare list shows the full catalog)
• MCP protocol support for external tool integration
• Security-hardened with SSRF protection, path traversal prevention, and resource limits
• On-device LLM inference support (1B-8B models)

Prerequisites

• aflare CLI installed (go install github.com/alib8b8/aflare/cmd/aflare@latest)
• Go 1.21+ or a pre-built binary
• Optional: Ollama for local LLM inference

Instructions

1. Generate a workflow: Describe the task in natural language
2. Review the YAML: The generated workflow is deterministic and editable
3. Execute: Run the workflow with progress tracking
4. Chain outputs: Use {{step.<name>}} (or {{step.<N>}}, 0-based index) to pass data between steps

CLI Commands

# Generate a workflow from plain English
aflare create "fetch weather data and save to file"

# Run a workflow file
aflare run workflow.yaml

# List all available nodes
aflare list

# Validate a workflow without executing
aflare validate workflow.yaml

# Dry run (show steps without executing)
aflare --dry-run run workflow.yaml

# Safe mode (disables execute node)
aflare --safe-mode run workflow.yaml

# Register a named connector (database / files / notes)
aflare connector add my-notes --type notes --root ~/notes

# List / inspect / remove connectors
aflare connector list
aflare connector show my-notes
aflare connector remove my-notes

Available Nodes

Utility Nodes:

Connector-aware nodes: sql_query, file_read, file_write, files_list accept a connector param — reference a named connector (registered via aflare connector add) instead of inline DSNs/paths. Credentials stay out of workflow files; read-only by default; connector ceilings (max_rows / timeout / max_bytes) always apply.

LLM Nodes:

Control Nodes:

Output

The skill generates a YAML workflow file and optionally executes it. Output includes:

• Generated workflow YAML
• Execution progress and logs
• Final output from the last step
• Error details if any step fails

Examples

Example 1: Fetch and Save

name: fetch-and-save
description: Fetch data from API and save to file
steps:
  - node: fetch_url
    name: fetch
    params:
      url: "https://api.example.com/data"
  - node: file_write
    params:
      path: "data.txt"
      content: "{{step.fetch}}"

Example 2: Fetch, Parse, and Summarize

name: summarize-article
description: Fetch article and summarize with LLM
steps:
  - node: fetch_url
    name: fetch
    params:
      url: "https://example.com/article"
  - node: ollama
    name: summarize
    params:
      model: "llama3"
      prompt: "Summarize: {{step.fetch}}"
  - node: file_write
    params:
      path: "summary.md"
      content: "{{step.summarize}}"

Example 3: Multi-source Aggregation

name: aggregate-sources
description: Fetch from multiple APIs and analyze
steps:
  - parallel:
      - node: fetch_url
        params:
          url: "https://api1.example.com"
      - node: fetch_url
        params:
          url: "https://api2.example.com"
  - node: combine
    name: merged
    params:
      separator: "\n"
  - node: ollama
    params:
      model: "llama3"
      prompt: "Analyze: {{step.merged}}"

Resources

• Full node catalog & parameters: see nodes-reference.md
• Ready-to-use workflow examples: see examples.md
• GitHub: https://github.com/alib8b8/aflare
• Documentation: https://github.com/alib8b8/aflare/blob/main/README.md
• Issues: https://github.com/alib8b8/aflare/issues
• License: AGPL-3.0