Skip to content

README_APPS.md

Latest commit

 

History

History
112 lines (76 loc) · 3.57 KB

README_APPS.md

File metadata and controls

112 lines (76 loc) · 3.57 KB

CopilotForge Apps Platform (Product Direction)

This document captures the next product direction for CopilotForge:

  • Move from generic “workflow” framing to Apps for non-technical users
  • Keep technical complexity hidden behind AI and guardrails and agent Skills folder.
  • Use the same Copilot brain/tooling stack under the hood, Native tools and Skills or custom one if needed.

It is intentionally product-first and not tied to a specific implementation.

Core Concept

Each app starts from a trusted Base Template and can be customized by AI for each user.

  • Base Template (owned by platform):
    • required inputs/form schema
    • default system prompt and tool permissions
    • output shape and quality checks
    • non-editable guardrails
    • an agent skill folder for that app. agent aksill gets update absed on user custimsation.
    • Custom native tool and agent tool if required for those flows
  • User Customization (owned by user):
    • tone, audience, format preferences
    • app-specific behavior adjustments through natural language
    • “make this app work like X for my use case”

This avoids exposing technical abstractions like tools, dependencies, or prompt internals.

Non-Technical UX Model

Simple flow:

  1. Pick an app from catalog
  2. Enter inputs in form
  3. Generate output
  4. Copy/export result

For app editing:

  • Preview: draft edits + test runs
  • Live: currently published app behavior
  • Publish: promote preview revision to live
  • Rollback: restore prior live revision

All internals remain hidden.

Versioning Model

Every app change is versioned.

  • Immutable revision history
  • Live pointer to a published revision
  • Preview pointer to an editable draft revision
  • One-click rollback to previous live revision

This keeps app evolution safe for non-technical users.

Sidecar Tool-Library Direction

Sidecar should expose a discoverable local tool library so AI can compose functionality without requiring users to install dependencies manually.

Goals:

  • AI discovers available capabilities from sidecar
  • Stable tool contracts (input/output schema)
  • Guarded execution with allowlists and safety checks
  • Clear execution traces for debugging and trust

Models and Specialized Capabilities

Default path: Copilot-backed model for most app tasks.

Optional routed capabilities:

  • image generation via Replicate (for apps that need visual assets)
  • specialized summarization/retrieval paths when app requires it

Users should see simple quality/speed choices, not model internals.

Monetization Split (Current Direction)

  • Mac app: free core experience
  • iOS Companion: Pro-only
  • Advanced capabilities (higher limits, premium app packs, image generation): Pro

This preserves a low-friction entry while funding advanced features.

Example Apps

  • Social Media Post Builder
    • inputs: topic, URL, resources, target platforms, post count
    • outputs: channel-specific post drafts + copy blocks
  • News Aggregator Brief
    • inputs: sources, timeframe, audience, style
    • outputs: structured digest cards + copy-ready summary

Guardrails and Trust

Guardrails are first-class and non-optional:

  • schema validation before publish
  • safety/compliance checks per app type
  • source/citation policy for research/news apps
  • safe fallback behavior when tools/models fail

Near-Term Product Milestones

  1. Define app template schema (inputs, outputs, guardrails, tool scope)
  2. Add preview/live and revision timeline UX
  3. Build AI customization flow for app variants
  4. Integrate sidecar tool-library discovery
  5. Add iOS companion access policy for Pro users