openmail/ARCHITECTURE.md

OpenMail — Architecture

Agent-native, self-hosted mail server. One Go binary that embeds Mox's (MIT) mail internals for the correctness-critical 95% — DKIM, SPF/DMARC, DANE + MTA-STS delivery, real-world MIME, spam filtering — and owns a native, agent-shaped data model (Postgres + object store) with a REST API and MCP server on top. This document is the design of record. It favors an MVP that actually receives and sends real mail over a perfect feature set.

0. The decision: embed Mox, don't rebuild and don't just wrap it

An agent-mailbox API is ~5% of the work. The other 95% is mail infrastructure (inbound :25 hardening, outbound deliverability, MIME, threading, quote-stripping, anti-abuse). Three ways to get that 95%:

1. Build from scratch — re-implement DKIM/SPF/DMARC/DANE/MTA-STS/MIME ourselves. Correct but slow, and we'd do it worse than a production server at first.
2. Sidecar an existing server (run Mox as a separate process, read it over IMAP). Fast, but we don't own the data model — threads/labels/extracted_text would be re-derived from IMAP, and we ship two coupled processes.
3. Embed Mox's packages into our own binary and own the storage + API. ← chosen.

Why (3) is viable: Mox has no internal/ directory — every package is importable — and the README states "most non-server Go packages are written to be reusable." The real boundary is not visibility, it's storage coupling: Mox's server + store packages are welded to its embedded bstore/bbolt store and global config; its protocol/crypto/parsing packages are stateless and clean. We embed the latter and replace the former with our own Postgres-backed server.

What we embed from Mox vs. what we own

Concern	Source	Notes
MIME parsing	mox message	Battle-tested on real, broken senders. Better than enmime for our needs.
DKIM sign + verify	mox dkim
SPF / DMARC	mox spf, dmarc	Inbound auth verification + policy.
Secure outbound delivery	mox smtpclient, dane, mtasts, dns	The deliverability moat — DANE/MTA-STS MX delivery, mostly solved by reusing this.
Bounces (DSN)	mox dsn	Parse/generate delivery status notifications.
Anti-abuse	mox iprev, dnsbl, ratelimit	rDNS check, blocklists, rate limiting on :25.
Spam filtering	mox junk	Built-in Bayesian filter; per-inbox train/score.
SMTP auth	mox sasl, scram	For the submission (send) path.
—	—	—
Inbound SMTP server loop	emersion/go-smtp	Clean, embeddable listener/session. We do NOT import mox smtpserver (store-coupled).
Storage	OpenMail (Postgres + S3)	Native agent data model. NOT mox store/bbolt.
API, MCP, threading, extraction, webhooks, auth, multi-tenancy	OpenMail	The agent-native layer — our differentiation.

Boundary rule: never import smtpserver, imapserver, queue, store, or mox- (config) — they assume Mox's storage/config. Importing them is the line between "embed as a library" and "fork Mox." We stay on the library side.

1. Component overview

                         ┌──────────────────────────────────────────────────────┐
   Internet senders ───▶ │  smtpd  (go-smtp on :25)                              │
   (MX → this VPS)       │   iprev/dnsbl/ratelimit → SPF/DKIM/DMARC verify       │
                         │   → mox/message parse → mox/junk score → enqueue      │   ← mox pkgs
                         └───────────────┬──────────────────────────────────────┘
                                         │ internal queue (Postgres LISTEN/NOTIFY)
                         ┌───────────────▼──────────────────────────────────────┐
                         │  core  (Go)                                            │
   Agents / SDKs ──REST──▶   • inbox/message/thread/draft services               │
   MCP clients ───MCP──▶ │   • threading, extracted_text, labels, FTS search     │
                         │   • auth (API keys, pods), webhooks/WS                 │
                         └──────┬───────────────────────────────┬────────────────┘
                                │                                │
                   ┌────────────▼─────┐            ┌─────────────▼──────────────┐
                   │ Postgres         │            │ Object store (S3/MinIO)    │
                   │ metadata + FTS   │            │ raw .eml + attachments     │
                   └──────────────────┘            └────────────────────────────┘
                                         ▲
                         ┌───────────────┴──────────────────────────────────────┐
   Outbound to world ◀── │  sender  (mox/dkim sign → backend)                    │
                         │   self-host: mox/smtpclient + dane + mtasts (MX)      │   ← mox pkgs
                         │   relay:     SES / Postmark / Resend API              │
                         └──────────────────────────────────────────────────────┘

All roles (smtpd, core API, sender, mcp) ship in one binary with subcommands; deploy together or split later. Caddy terminates TLS for the HTTP API and auto-manages certs.

2. Tech stack

• Language: Go — single static binary, and the whole mature self-hosted-mail ecosystem (Mox, go-smtp, go-message) is Go, keeping us MIT-compatible. Rust (Stalwart) is AGPL → ruled out.
• HTTP: chi router; sqlc + pgx for typed Postgres access.
• Inbound server: emersion/go-smtp listener → handed to mox packages for verify/parse/score.
• Mail correctness/delivery: mox message, dkim, spf, dmarc, dane, mtasts, dns, smtpclient, dsn, junk, iprev, dnsbl, ratelimit, sasl, scram (see §0 table).
• Quote stripping: a talon/EmailReplyParser-style heuristic pass for extracted_text.
• Storage: Postgres (metadata + tsvector FTS); S3-compatible object store (MinIO self-host, or S3) for raw .eml and attachment blobs.
• Events: Postgres LISTEN/NOTIFY for the internal inbound queue and to fan out webhooks + a WebSocket stream. (Swap to NATS only if volume warrants.)
• Deploy: docker compose — caddy, openmail (one image, multiple commands), postgres, minio. MX + SPF/DKIM/DMARC/DANE/MTA-STS DNS documented in deploy/.

3. Data model (Postgres)

Native and agent-shaped (not Mox's per-account bbolt index). Resource shapes mirror what agent tooling expects so SDKs/MCP map cleanly.

pods          (id, name, created_at)                       -- tenant isolation
inboxes       (id, pod_id, address UNIQUE, display_name, dkim_selector, metadata jsonb, created_at, updated_at)
threads       (id, inbox_id, subject, last_message_id, message_count, labels text[], updated_at, created_at)
messages      (id, inbox_id, thread_id, message_id_hdr, in_reply_to, references text[],
               from_addr, to_addrs text[], cc text[], bcc text[], subject,
               preview, text, html, extracted_text, extracted_html,
               raw_object_key,            -- pointer to raw .eml in object store
               spf, dkim, dmarc,          -- inbound auth verdicts (from mox pkgs)
               junk_score, labels text[], size_bytes, headers jsonb, ts tsvector,  -- FTS
               created_at, updated_at)
attachments   (id, message_id, filename, content_type, size_bytes, object_key, inline bool, content_id)
drafts        (id, inbox_id, thread_id, to_addrs text[], cc, bcc, subject, text, html, send_at, client_id)
api_keys      (id, pod_id, hash, scopes text[], created_at)   -- bearer tokens; store only the hash
webhooks      (id, pod_id, url, event_types text[], secret)   -- HMAC-signed deliveries
outbox        (id, message_id, status, attempts, last_error, next_attempt_at)  -- send queue + retries
events        (id, inbox_id, type, payload jsonb, created_at) -- audit + WS replay
domains       (id, pod_id, name, dkim_privkey_ref, verified bool)  -- per-domain DKIM keys + DNS state

Indexes: inboxes(address), messages(thread_id), messages(inbox_id, created_at desc), GIN on messages.ts (search) and *.labels. Raw MIME + attachments live in object storage; rows hold pointers.

4. Request / data flows

Inbound (receive)

1. go-smtp accepts on :25; anti-abuse gate via mox iprev/dnsbl/ratelimit + size cap; raw .eml written to object store.
2. mox spf/dkim/dmarc verify the message; verdicts recorded. mox message builds the MIME tree → text/html, attachments, headers. mox junk scores spam. Quote-stripper computes extracted_text + preview.
3. Threading resolves thread_id from In-Reply-To/References, falling back to normalized-subject
   • participants within a window.
4. Row inserted; LISTEN/NOTIFY fires → message.received webhook + WebSocket event.

Outbound (send / reply)

1. API builds the RFC 5322 message (sets In-Reply-To/References on replies); persists a messages row + outbox entry.
2. sender mox-dkim-signs, then dispatches via the configured backend:
   • self-host: mox smtpclient + dane + mtasts for authenticated, secure MX delivery.
   • relay: SES/Postmark/Resend API.
3. Bounces/complaints (mox dsn parse, or relay webhook) update outbox.status and emit events.

API surface (v1 — AgentMail-shaped)

POST   /v0/inboxes                         create inbox
GET    /v0/inboxes  /  /v0/inboxes/{id}    list / get
POST   /v0/inboxes/{id}/messages/send      send
GET    /v0/inboxes/{id}/messages           list (limit, page_token, labels)
GET    /v0/inboxes/{id}/messages/{id}      get
POST   /v0/inboxes/{id}/messages/{id}/reply
GET    /v0/inboxes/{id}/threads  /  /{id}  list / get
…drafts, webhooks, search…

Bearer auth (Authorization: Bearer). Keeping paths/shapes close to AgentMail's public v0 lets an existing client (or MCP server) target a self-hosted OpenMail by base-URL swap. Compatibility yields to a cleaner native shape where they conflict.

MCP server

A thin MCP front-end over core: tools create_inbox, list_messages, get_thread, send_message, reply, search — so an agent drives its own mailbox directly.

5. Deliverability plan (the moat — and why embedding Mox helps most here)

• Self-host send is achievable from day one for the protocol parts because mox smtpclient + dane + mtasts + dkim already implement authenticated, DANE/MTA-STS-secured MX delivery. What remains is operational, not code: dedicated IP with PTR/rDNS, SPF (v=spf1 ip4:… -all), published DKIM key, DMARC (p=quarantine→p=reject), gradual IP warmup, FBL enrollment, suppression lists.
• Relay backend (SES/Postmark/Resend) ships alongside for instant inbox placement while a self-host IP warms up — chosen per-domain.
• Inbound auth: SPF/DKIM/DMARC verified on receipt (mox pkgs); verdicts exposed in message metadata; optional reject/quarantine policy per inbox.

6. MVP milestones

1. Core API + storage — inboxes/messages/threads/drafts CRUD on Postgres + MinIO; bearer auth; ingest endpoint to seed without real mail. Proves the data model + API.
2. Inbound — go-smtp on :25 + mox verify/parse/junk + threading + message.received webhook. MX a test domain at the VPS. Proves receive.
3. Outbound via relay — send/reply, mox-DKIM-signed, through a relay. Proves the loop: an agent receives and replies.
4. MCP server — agent owns and operates an inbox end-to-end.
5. Self-host SMTP send — wire mox smtpclient/dane/mtasts; the deliverability/warmup long tail.

7. Repo layout (planned)

cmd/openmail/         # single binary: `serve`, `smtpd`, `sender`, `mcp` subcommands
internal/core/        # inbox/message/thread/draft services
internal/smtp/        # go-smtp inbound server; calls mox verify/parse/junk + threading
internal/sender/      # Sender interface: smtp (mox smtpclient+dane+mtasts) & relay backends; mox dkim
internal/store/       # sqlc queries, object-store client
internal/api/         # chi HTTP handlers (v0 surface)
internal/mcp/         # MCP server over core
deploy/               # docker-compose.yml, Caddyfile, DNS (MX/SPF/DKIM/DMARC/DANE/MTA-STS) notes
db/migrations/        # SQL migrations

8. Licensing & attribution

OpenMail is MIT (see LICENSE). Every dependency is permissive: Mox (MIT), emersion/go-smtp and go-message (MIT), pgx/chi (MIT/BSD). We import Mox as a Go module (no vendored source in this repo), but credit it prominently in the README; if we ever vendor Mox source we must retain its MIT notice. No GPL/AGPL code anywhere — Maddy (GPL-3.0) and Stalwart (AGPL) are deliberately avoided so the project stays MIT.

9. Open questions

• mox message ⇄ go-smtp seam: confirm mox's parser consumes a raw reader cleanly outside mox's store (spike in milestone 2). If any Tier-1 package drags in mox-/store transitively, isolate behind a thin adapter or vendor just that file.
• How far to chase AgentMail API compatibility vs. a cleaner native shape.
• Search: Postgres FTS is enough for v1; revisit (Meilisearch/Typesense) only if needed.
• Multi-tenancy depth — pods now, full RBAC later.
• Whether to expose an optional IMAP read path (via mox imapserver over our store would mean forking it — more likely a thin standalone IMAP front-end later, kept out of v1).