diff --git a/docs/deploy-dokploy.md b/docs/deploy-dokploy.md
new file mode 100644
index 0000000..6beb174
--- /dev/null
+++ b/docs/deploy-dokploy.md
@@ -0,0 +1,77 @@
+# Self-Hosting on Dokploy
+
+This guide covers deploying OpenReply on your own server using [Dokploy](https://dokploy.com), as an alternative to the Vercel + Railway setup covered in `docs/setup.md`. Running everything on your own Dokploy instance means no per-seat hosting fees and no usage caps — but there are a few gotchas specific to this setup worth knowing up front.
+
+## Overview
+
+OpenReply needs four services running:
+- **Web app** — the Next.js dashboard, OAuth callback, and webhook receiver
+- **Worker** — a long-running Node process that sends the DMs (cannot run as a serverless function)
+- **Postgres**
+- **Redis**
+
+On Dokploy, these become: one Postgres database service, one Redis database service, and **two separate Applications** pointing at the same repo (one for the web app, one for the worker), each with different start commands.
+
+Since everything runs on the same Dokploy server, you don't need the "public vs internal database URL" split that the Vercel/Railway guide requires — both apps can just use Dokploy's internal service hostnames directly.
+
+## Gotcha #1 — Node version
+
+The repo doesn't pin a Node version, so Nixpacks (Dokploy's default build system) will default to an old version — too old for this project's dependencies (Prisma 7 and Next.js 16 both require Node 20.19+, 22.12+, or 24+).
+
+Depending on which Nixpacks version your Dokploy instance ships, you may also find:
+- Requesting Node 22 via `NIXPACKS_NODE_VERSION=22` resolves to a patch (e.g. 22.11.0) just below Prisma's 22.12 minimum
+- Requesting Node 24 fails outright with `undefined variable 'nodejs_24'`, if the pinned Nix package snapshot predates Node 24's availability
+
+**Fix:** add a `nixpacks.toml` file to the repo root to pin a newer, known-good Nix package snapshot:
+
+```toml
+[phases.setup]
+nixpkgsArchive = "51ad838b03a05b1de6f9f2a0fffecee64a9788ee"
+```
+
+This snapshot provides Node 22.13.1, which satisfies Prisma's requirement. Then set:
+
+```
+NIXPACKS_NODE_VERSION=22
+```
+
+as an environment variable on both the web app and worker app in Dokploy.
+
+## Gotcha #2 — No dedicated "Build Command" field
+
+Unlike Vercel/Railway, Dokploy's Nixpacks builder reads `package.json` scripts automatically and has no separate Build Command / Start Command UI field. To override the detected commands, set these as **environment variables** instead:
+
+**Web app:**
+```
+NIXPACKS_BUILD_CMD=npx prisma generate && next build
+NIXPACKS_START_CMD=npx prisma migrate deploy && npm start
+```
+
+**Worker app:**
+```
+NIXPACKS_BUILD_CMD=npm run db:generate
+NIXPACKS_START_CMD=npm run worker
+```
+
+## Gotcha #3 — Migrations must run at start, not build
+
+This is the one most likely to trip people up: **`prisma migrate deploy` cannot run during the Docker build step.** Docker builds run in an isolated environment with no access to Dokploy's internal service network, so the database isn't reachable yet — you'll see:
+
+```
+Error: P1001: Can't reach database server at `<db-service-name>:5432`
+```
+
+if you try to run migrations as part of the build (e.g. via the `vercel-build` script, which bundles `prisma generate && prisma migrate deploy && next build` together). The fix is the build/start command split shown above — `prisma generate` and `next build` happen at build time (no DB needed), while `prisma migrate deploy` runs in the start command, once the container is actually live and on the network.
+
+## Step-by-step
+
+1. Fork this repo.
+2. In Dokploy: **Create → Database → PostgreSQL** and **Create → Database → Redis**. Note their internal service hostnames.
+3. In Dokploy: **Create → Application**, connect your fork, `main` branch. This is the web app.
+4. Add the `nixpacks.toml` file (Gotcha #1) to your fork's root, and set the environment variables from Gotchas #1 and #2 (web app version) plus the standard variables from `.env.example` — pointing `DATABASE_URL` and `REDIS_URL` at the internal hostnames from step 2.
+5. Assign a domain to the web app only (not the worker) in Dokploy's Domains section, container port `3000`. This becomes your `NEXTAUTH_URL`.
+6. Repeat step 3 for a second Application — this is the worker. Use the worker's build/start commands from Gotcha #2, and the same full set of environment variables as the web app, especially `DATABASE_URL`, `REDIS_URL`, and `ENCRYPTION_KEY` (these three must match exactly between both apps, or DM sends will fail to decrypt).
+7. Deploy both apps.
+8. Check `https://your-domain/api/health` — confirms database, Redis, queue, and worker heartbeat are all healthy.
+
+From here, the Meta app setup, OAuth redirect, and webhook configuration are identical to the standard setup in `docs/setup.md`.