iiinova — where your digital self keeps evolving

DeployHQ @deployhq.com · 15/07/2026

Full walkthrough — the DSN, the MCP config, and the cp-based safety net — here: www.deployhq.com/blog/using-c...

DeployHQ @deployhq.com · 15/07/2026

Two habits that keep it honest: • Back up before writes, not reads — selects can't hurt the file • Read the SQL before you run it — the assistant is drafting, not deciding Work out the shape where mistakes are free, then take the validated result to a real server.

100

DeployHQ @deployhq.com · 15/07/2026

If it goes sideways: cp mydb.backup.sqlite mydb.sqlite You're back byte for byte. No migration to reverse, no pg_restore. The safety net is the filesystem itself. So you can be reckless on purpose: try aggressive denormalization, rewrite five tables, compare three backfills.

100

DeployHQ @deployhq.com · 15/07/2026

Here's the trick nobody documents: your entire database is one file, so your rollback plan is one command. cp mydb.sqlite mydb.backup.sqlite Before any ALTER, bulk UPDATE, or schema rewrite the assistant suggests — copy the file first.

100

DeployHQ @deployhq.com · 15/07/2026

SQLite is that database. No server, no port, no user to grant. The whole thing is one .sqlite file you can open, copy, delete, and restore in seconds. Point Claude Code at it (via the DBHub MCP server) and let it read your real schema — not a hallucinated one.

110

DeployHQ @deployhq.com · 15/07/2026

Every guide on giving an AI assistant database access starts with the hard case: production Postgres, credentials, network access, read-only roles. That's backwards. The safest place to start is a database that lives in one file on your laptop. 🧵

100

DeployHQ @deployhq.com · 07/07/2026

6/ Full write-up — the GRANT SQL for Postgres/MySQL/SQL Server, the MCP flags, the sandbox pattern, and a copy-paste checklist: www.deployhq.com/blog/ai-data...

000

DeployHQ @deployhq.com · 07/07/2026

5/ Step three: for anything that needs to write, give the AI a disposable copy — a throwaway Postgres loaded from a sanitized dump. It can mutate all it wants; nothing real is at risk. Schema changes go through your deploy pipeline, never the agent.

100

DeployHQ @deployhq.com · 07/07/2026

4/ Step two: guardrails at the MCP layer. `--readonly` and `--max-rows` flags stop a "quick look" from turning into a full-table scan or a mutation. Defense in depth — the role is the wall, the flags are the moat.

100

DeployHQ @deployhq.com · 07/07/2026

3/ Step one is a dedicated read-only role. Not your app user. Not a shared credential. A role with `GRANT SELECT` and nothing else — no INSERT, UPDATE, DELETE, or DDL. Postgres, MySQL, and SQL Server each have a one-time setup for this.

100

DeployHQ @deployhq.com · 07/07/2026

2/ The AI didn't malfunction. It generated a plausible-looking command and executed it — exactly what it's designed to do. If that command can DROP a table, the problem is the grant, not the model.

100

DeployHQ @deployhq.com · 07/07/2026

1/ "An AI assistant deleted our production database" isn't a story about a rogue AI. It's a story about a connection string with write permissions and no seatbelt. The fix is boring and it works: read-only guardrails. 🧵

100

DeployHQ @deployhq.com · 24/06/2026

Full walkthrough (incl. OAuth redirect, RLS context, CORS gotchas): www.deployhq.com/blog/deploy-... Trial site is free.

010

DeployHQ @deployhq.com · 24/06/2026

Migration is mostly: – GitHub sync (Lovable built-in) – Static Hosting site provisioned in DeployHQ – VITE_SUPABASE_* env vars – Supabase Auth redirect URL allowlist – Optional CNAME for a custom domain Auth/DB/storage/Edge Functions all stay on Supabase.

110

DeployHQ @deployhq.com · 24/06/2026

Lovable's output is a React+Vite bundle — a flat directory of static files. That's a fit for any static-hosting layer. If you're already on Cloudflare DNS or used to Cloudflare Pages, Cloudflare's edge under DeployHQ Static Hosting is the closest match. No credit metering.

100

DeployHQ @deployhq.com · 24/06/2026

The shape: sync project to GitHub, keep Supabase doing the backend work, continue in your own editor with Claude Code or Cursor, host the static frontend somewhere predictable. Last bullet usually defaults to Vercel. It doesn't have to.

110

DeployHQ @deployhq.com · 24/06/2026

Lovable's good at the scaffold. Less good at the third iteration on the same screen — that's where most teams hit a credit wall and start looking for what's next. The community playbook for moving past it is mostly one shape. Sharing the variant we just wrote up.

100

DeployHQ @deployhq.com · 16/06/2026

Pick the camp before the tool: – Camp 1: predictable, locked in – Camp 2: swap models, cap spend – Camp 3: flat bill, OSS only All 7 compared: www.deployhq.com/blog/claude-...

010

DeployHQ @deployhq.com · 16/06/2026

The agents + IDEs that plug into the routers: • Aider — terminal, git-native • Cline — autonomous IDE agent, MCP-native • Continue.dev — Apache 2.0, VS Code + JetBrains • Cursor — single-vendor, hybrid pricing Mix-and-match is the point.

110

DeployHQ @deployhq.com · 16/06/2026

3/ Flat-rate + curated OSS. OpenCode: MIT-licensed, 160k+ GitHub stars, 900+ contributors. Optional flat-rate Go subscription for curated Chinese OSS models. One bill, no metering. Ceiling below frontier, but solid for daily coding.

110

DeployHQ @deployhq.com · 16/06/2026

2/ Portable + metered. OpenRouter (hosted) or LiteLLM (self-hosted) sit between you and the model providers. 400+ models, one key, swap freely. LiteLLM adds virtual keys per engineer + hard spend caps. The runaway-bill insurance policy.

110

DeployHQ @deployhq.com · 16/06/2026

1/ Locked-in + metered. Claude Code, Copilot, Cursor. Subscription + per-token overages once you blow past the included usage. Polished. Predictable until it isn't. Your workflow stays with the vendor — no easy exit.

320

DeployHQ @deployhq.com · 16/06/2026

The AI coding market is splitting into 3 pricing camps, not consolidating. 1/ Locked-in + metered (Claude Code, Copilot, Cursor) 2/ Portable + metered (OpenRouter, LiteLLM) 3/ Flat-rate + curated OSS (OpenCode) Quick breakdown of each ↓

220

Reposted by DeployHQ

Honeybadger.io @honeybadger.io · 20/05/2026

With DeployHQ’s new integration, every deploy gets pushed into Honeybadger with its commit data, so when something breaks, you’ll see what changed, when it shipped, and who shipped it. Here’s how to use it → www.deployhq.com/blog/honeyba... @deployhq.com