Scope & Limitations

What Ginie handles well

• Single-module Daml contracts of 50–200 lines
• Bilateral and 2–3 party workflows
• Standard financial contract patterns: IOU, bond, repo, DvP, custody, AML, settlement
• Standard choice structures: transfer, settle, archive, propose/accept
• SCU-safe contracts designed for on-chain upgrade from day one
• Full project export — contracts you can own, maintain, and extend independently
• Programmatic generation via Python SDK for batch and CI use cases

What Ginie is not

• Not a cloud IDE or terminal emulator
• Not a Daml language assistant or documentation chatbot
• Not a replacement for Daml Autopilot (which serves existing Daml developers)
• Not a replacement for splice or multi-module enterprise Daml packages
• Not a smart contract security auditor for human-authored Daml (it audits its own output)

Comparison: Ginie vs manual Daml development

FAQ

Do I need a Canton account or credentials?

No. GinieNet is a public Canton sandbox — you can deploy contracts without any credentials. For deploying to Canton Mainnet or your own LocalNet, you will need the appropriate Canton party credentials.

What LLM does Ginie use?

Claude Sonnet is the default — it produces the highest first-pass Daml compile rates in Ginie's production benchmark. You can switch to GPT-4o, Gemini 2.5 Flash, or (from M4) Ginie-1 using the llm parameter in the SDK or the model selector in the web UI.

Is my contract source code stored?

Generated Daml source and audit reports are stored temporarily for the duration of a session. With the M2 GitHub integration, your code is committed to a repository you own and control. You can delete your data via the dashboard at any time.

Can I use Ginie with my own Canton node?

Yes. Set CANTON_LEDGER_HOST in your environment to point at your own Canton participant node. The deploy agent will target your node instead of GinieNet. Fully supported in self-hosted deployments and the Python SDK.

What happens if compilation fails after 3 retries?

The pipeline returns the final compiler error alongside the generated Daml source so you can inspect what went wrong. You can then iterate with a more specific prompt or modify the source directly. In most cases, rephrasing with more explicit party and choice structure resolves the failure.

Can I contribute Daml patterns to the RAG library?

Yes. Submit a PR to Canton Skills. Every merged pattern is validated against Canton SDK 3.4 by CI before entering the corpus. Merged patterns automatically propagate to Ginie's RAG layer on the next build.

Is there a rate limit on GinieNet?

GinieNet is a shared public sandbox. During the public preview period, there is a soft limit of 10 deployments per party per day. For higher volume usage, run the self-hosted version or contact ginie@blockx.ai.

What is Canton Skills and how does it differ from Ginie?

Canton Skills is a standalone npm package (npx canton-skills install) that provides a structured Canton knowledge base for AI coding agents. Ginie uses Canton Skills internally as its primary knowledge layer. Canton Skills works with any AI editor — Claude Code, Cursor, Copilot — not just Ginie.

How does Ginie handle existing deployed contracts?

You can iterate on a previously generated contract using the Iterate button in the UI or the client.iterate() SDK method. The SCU Upgrade Agent (M2) handles upgrade lifecycle management for contracts you want to modify on-chain.