Files
hermes-mobile/docs/DIRECT_GATEWAY_ARCHITECTURE.md
T
2026-07-24 13:31:37 +00:00

112 lines
8.9 KiB
Markdown

# Hermes Mobile Direct Gateway Architecture
Hermes Mobile is pivoting to a native Android client that talks directly to a remote upstream Hermes Agent gateway/API server. The `apps/companion` TypeScript server is legacy compatibility code and must not be required by Android or future releases.
## Architecture
```text
Android app
Kotlin/Compose UI
connection/auth storage
transport DTOs
domain repositories
|
| HTTPS, bearer token when configured
v
Upstream Hermes Agent gateway/API server
auth/session authority
chat/completion authority
streaming/tool/artifact authority
provider credentials and Hermes runtime
```
The Android app remains a thin client. It may store server profiles, non-provider UI preferences, selected session IDs, and Android-secured bearer tokens. It must not embed provider credentials, clone Hermes backend behavior, or make the legacy companion server part of the release path.
## Documented Upstream Surface
The current repository/context documents these upstream Hermes API-server endpoints:
| Endpoint | Status | Mobile contract |
| --- | --- | --- |
| `GET /health` | Documented | Primary reachability and compatibility probe. Response shape beyond a successful JSON response is treated as server-specific. |
| `GET /v1/models` | Documented | Authenticated compatibility probe for bearer-token connections and future model picker source. |
| `GET /v1/capabilities` | Documented | Capability negotiation source. Current upstream behavior advertises endpoints as `{method, path}` objects and feature support as booleans. |
| `POST /v1/chat/completions` | Documented | Baseline OpenAI-compatible chat path. Mobile must assume this may not expose Hermes session IDs or structured tool events. |
| `POST /v1/responses` | Documented | Preferred future request/streaming path if upstream exposes richer response events. Exact Hermes event mapping is not yet verified. |
| `/api/sessions/*` | Documented and capability-advertised | List/create/get/update/delete/messages/fork/chat/chat-stream routes exist, but mobile must use only the exact advertised method/path object and must not infer search or other routes. |
| `POST /api/sessions/{session_id}/chat/stream` | Documented and capability-advertised | SSE session turn. Core mobile support begins with `message.started`, `assistant.delta`, authoritative `assistant.completed`, `error`, and `done`; tool events remain gated by `tool_progress_events`. |
Any route not listed here must be labeled provisional until checked against official Hermes Agent docs or a running gateway. Session search remains explicitly unverified and unmapped.
## Direct stream lease contract
- Parse `session_chat_streaming`, `tool_progress_events`, and `endpoints.session_chat_stream` from `GET /v1/capabilities`.
- Require the advertised session stream endpoint to be `POST`, use an absolute path containing `{session_id}`, and send the documented JSON `input` field.
- Bind the request URL, route generation, session ID, endpoint path, and structured-event gate into one immutable lease before the HTTP call starts.
- Execute that prebuilt request through OkHttp without consulting route selection again until the call ends.
- Never apply an SSE event whose `session_id` differs from the bound lease. A later Local/Remote re-evaluation applies only to a later request.
- Append `assistant.delta` text deterministically, then replace it with the authoritative `assistant.completed.content` response when present.
- Report HTTP, I/O, or malformed-SSE failure only against the route/session lease that owned the failed call; an upstream `error` event fails the session without invalidating the route.
- Decode and render tool lifecycle/progress only when `tool_progress_events` is advertised. Treat event payload fields defensively, retain large output outside bounded display windows, and persist only client presentation state such as expansion.
- Do not expose artifact share/save actions until an upstream artifact handoff capability and URI/download contract are advertised; the current direct gateway contract has neither.
## Unverified Assumptions
The following are useful design assumptions, not confirmed contracts:
- Bearer authentication is accepted through an `Authorization: Bearer <token>` header on `/v1/*`.
- `/health` returns a stable `status` field.
- `/v1/responses` can provide enough streaming metadata to preserve Hermes sessions, artifacts, cancellation, and reconnect state.
- Hermes session search has a stable direct HTTP endpoint.
- Tool result and artifact listing/download have stable direct HTTP endpoints.
- Cancellation has a stable direct HTTP endpoint.
- Error payloads consistently include `message` or `error`.
Android contracts may model these needs, but implementation must keep them capability-gated until verified.
## Compatibility Policy
- Direct gateway routes are the default target for all new Android transport work.
- Legacy companion routes such as `/api/health`, `/api/auth/validate`, `/api/chat`, `/api/files/*`, and `/api/terminal/*` are temporary compatibility only.
- UI and domain terminology should say "gateway", "server", or "upstream Hermes" unless explicitly describing legacy companion behavior.
- The transport/domain model exposes direct gateway capabilities first. `legacyCompanionAdapter` exists only to identify old adapters during migration.
- Backend behavior belongs upstream in Hermes Agent. Hermes Mobile may add client adapters and mappers, not server-side replacements.
## Local-first gateway routing
A server profile can contain two URLs for the **same Hermes gateway identity**:
- **Local URL** — preferred when the phone can safely reach the gateway on the home/LAN (for example, `https://hermes.home.arpa`).
- **Remote URL** — fallback when local reachability is unavailable (for example, a configured public HTTPS hostname).
The client must choose automatically per connection attempt:
1. If a local URL is configured, probe its documented `GET /health` endpoint using a short, bounded timeout.
2. Use the local URL when the probe succeeds and the gateway identity/auth contract is compatible.
3. Otherwise immediately use the configured remote URL; never leave chat blocked waiting for LAN discovery.
4. Re-evaluate on app foreground, explicit reconnect, connectivity changes, and after a failed active request. Do not switch a request mid-stream.
5. Show the active route (`Local` or `Remote`) in connection diagnostics, but do not expose bearer tokens or internal host details in message content/logs.
Security and reliability rules:
- Local and remote URLs are explicit user configuration, not guessed IP ranges or silently discovered devices in the MVP.
- Both routes must use HTTPS by default and must be treated as the same intended gateway only after an authenticated compatibility check. The app must not silently send a bearer token to an arbitrary captive portal or unrelated LAN host.
- Remote is a fallback, not a duplicate profile. User preferences, selected session state, and credentials stay associated with one logical gateway profile.
- A later milestone may add opt-in mDNS/QR setup, certificate pinning/identity binding, and richer connectivity observation only after direct-gateway basics are verified.
- The current upstream identity-binding limitation and the stronger contract needed are tracked in `docs/BLOCKERS.md`.
## Android Contract Milestones
1. Auth and connection: validate local + remote URLs, store bearer tokens securely, probe `GET /health`, use `GET /v1/models` as the first authenticated check, and make local-first fallback observable/retryable.
2. Sessions: keep repository interfaces for list/create/select/continue, but mark route mapping provisional until upstream session docs are verified.
3. Streaming: support final response and token delta reduction now; gate structured tool/artifact/session events behind server capabilities. Keep each stream bound to its selected local or remote route.
4. Companion removal: after direct gateway chat/session/streaming works end to end, remove legacy companion UI routes and then deprecate install/docs around `apps/companion`.
## Verified Session Capability Boundary — 2026-07-24
Official upstream API server documentation defines authenticated `GET /v1/capabilities` discovery and advertises session endpoint entries including `session_list`, `session_create`, `session_get`, and `session_delete`. The Android B3 foundation therefore builds requests only from those returned endpoint strings; a missing entry leaves that operation unavailable rather than falling back to an assumed route.
Source checked: `https://hermes-agent.nousresearch.com/docs/user-guide/features/api-server/` on 2026-07-24.
The same official endpoint table does not document session search or a search query schema. Hermes Mobile keeps search in the repository contract and fake implementation for UI/state verification, but does not map search to HTTP until upstream publishes that contract. The exact blocker is tracked in `docs/BLOCKERS.md`.