Competitive Analysis

Competitive Analysis is a local-first Tutti workspace app for product research. It is chat-first: you ask a local agent to research a product, and it runs an evidence-backed teardown that stays attached to the workspace.

What is included

• A monorepo aligned with the group-chat-style Tutti app architecture:
  • apps/web: React + Vite chat UI
  • apps/server: Fastify local runtime
  • packages/shared: shared domain contracts
• The vendored product-swipefile research skill under apps/server/skills/, injected into every agent run.
• Tutti packaging metadata: tutti.app.json, tutti.cli.json, bootstrap.sh, scripts/package-tutti-app.mjs
• Local-first persistence and Tutti references:
  • sessions, messages, and artifacts stored under TUTTI_APP_DATA_DIR/sessions
  • captured research artifacts exposed through /tutti/references/list and searchable via /tutti/references/search
• Ecosystem integration so other Tutti apps and agents can drive it:
  • a competition CLI scope (status, sessions, reports, report, research) declared in tutti.cli.json — see COMMANDS.md
  • graceful TUTTI_CLI consumption for calling sibling apps / the daemon
• Internationalization (English + 简体中文) with light/dark theming via prefers-color-scheme

Tutti ecosystem

Other apps and agents call this app through the bundled Tutti CLI:

"$TUTTI_CLI" --json competition status
"$TUTTI_CLI" --json competition reports --query notion
"$TUTTI_CLI" --json competition research --product "Linear"

See COMMANDS.md for the full command and HTTP surface.

How it works

1. Pick an available Agent Target from the current Tutti agent catalog.
2. Type a research request, e.g. 调研一下 Notion or research Cursor.
3. The agent runs the product-swipefile staged pipeline (evidence collection → inventory → gap checks → writing → validation) inside a per-run working directory, streaming its progress into the chat thread.
4. When it finishes, the produced report.md (plus inventory.md, meta.json, and the raw/ evidence cache) is captured into the session's artifact list and the unified research library, and surfaced back into Tutti as references.

ResearchRunService (orchestrator)
  -> LocalAgentResearchProvider  (@tutti-os/agent-acp-kit -> selected Agent Target)
      -> skillManifest materializes product-swipefile into the run cwd
      -> scanRunArtifacts captures report.md / inventory.md / meta.json

A compatible local Agent Target must be available and authenticated; there is no offline fallback. Discover current targets with tutti agent list --json.

Development

pnpm install
pnpm dev

The web app runs on Vite and proxies API calls to the local Fastify server.

Packaging

pnpm package:tutti

This writes a self-contained app package under build/tutti-app/package, including the bundled server, web assets, and the vendored skill.