Files
hapi/hub
927b4971b9 feat(dsh): model selection for DeepSeek Harness (ACP config options) (#2030)
* feat(acp): flatten Zed-style grouped configOptions choices

DeepSeek Harness 0.2.x (and Zed-style agents generally) nest select
choices one level under group entries that carry no `value` of their
own:

    options: [{ group, name, options: [{ value, name }] }]

Both the per-session config-option descriptor and the model-metadata
merge previously read only flat `{ value, name }` entries, so grouped
pickers surfaced zero choices. Flatten recursively in a shared helper;
flat options are unaffected.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* feat(dsh): runtime model switching through ACP config options

dsh advertises its model catalog as a Zed-style grouped config option
(provider groups with JSON-encoded provider+model tuple values) instead
of session/set_model, so wire it through the generic config-option rail:

- shared: ListDshModels/ListDshModelsForMachine RPC methods, response
  types, and grant dsh the ModelChange flavor capability
- cli: DshMode becomes {model} and rides the queued-batch mode like
  other flavors; the launcher applies model switches per-turn via
  session/set_config_option, discovers the live current model from the
  config option after newSession, rolls back (and warns) when a switch
  is rejected, and serves ListDshModels from session metadata. A
  machine-scoped probe (60s single-flight cache) backs the new-session
  catalog; --model is now accepted by the dsh command
- hub: dsh-models passthrough routes (machine + session scope)

Co-Authored-By: Claude Code <noreply@anthropic.com>

* feat(web): dsh model selector for new sessions and in-chat switching

New-session form gets a Default-plus-catalog picker fed by the machine
probe (opaque tuple values verbatim, friendly names as labels, stale
draft picks dropped when the catalog rotates). Running sessions get the
same catalog over the session-scoped hook; switching rides the existing
generic model-change path now that dsh advertises ModelChange.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* fix(dsh): honor explicit Default reset and cover pending catalog discovery

The launcher now captures the server default before applying any spawn
override, and a queued `model: null` (the picker's "Default" after an
explicit switch) switches back to it instead of silently keeping the
last applied model; DshMode keeps null distinct from undefined all the
way from the config RPC through the queue (runDsh no longer collapses
it). The web model-option helpers get explicit dsh branches mirroring
opencode: pending, failed, or empty ACP discovery renders no Claude
presets and the Ctrl/Cmd+M cycler keeps the current model. NewSession
restores dshSelectedModel from preferred launch settings alongside
agy/opencode (the existing catalog-validation effect drops stale tuples)
with remount, forward-on-create, and fallback tests.

Co-Authored-By: Claude Code <noreply@anthropic.com>

* fix(dsh): keep unset model preferences distinct from Default resets

Two review follow-ups:

- currentModel collapses "no preference yet" into null, so a session
  started without --model on a server that reports no current model
  queued model:null on every ordinary turn — the launcher read each one
  as an explicit "Default" reset and emitted 'Cannot reset the DSH
  model' every turn. The preference is now three-state (undefined =
  nothing requested, null = picker's explicit Default, string = model),
  encoded through dshModeForPreference() so preference-less turns carry
  no model field at all.

- ListDshModels was registered only after initialize/session/new and
  discovery finished; a picker opened during startup hit an
  unregistered-method rejection with no retry path and stayed empty
  until remounted. The handler is now registered before initialization
  and parks on a readiness promise resolved once the session exists and
  discovery has settled (rejected with the startup error otherwise).

Co-Authored-By: Claude Code <noreply@anthropic.com>

* docs: DSH advertises model selection over ACP

The capability blurbs still said HAPI exposes no model pickers for DSH and
that the ACP server has no model switching. Update them to the wire reality:
models are listed via ListDshModels and applied per turn with
session/set_config_option. Resume remains out of scope on this branch.

via [HAPI](https://hapi.run)

Co-Authored-By: HAPI <noreply@hapi.run>
Co-Authored-By: Claude Code <noreply@anthropic.com>

---------

Co-authored-by: Claude Code <noreply@anthropic.com>
Co-authored-by: HAPI <noreply@hapi.run>
2026-10-26 10:02:51 +08:00
..
2026-08-25 13:31:59 +08:00

hapi-hub

Telegram bot + HTTP API + realtime updates for hapi hub.

What it does

  • Telegram bot for notifications and the Mini App entrypoint.
  • HTTP API for sessions, messages, permissions, machines, and files.
  • Server-Sent Events stream for live updates in web and native clients.
  • Socket.IO channel for CLI connections.
  • Serves the web app from web/dist or embedded assets in the single binary; network-relay mode uses the separately hosted official web app.
  • Persists state in SQLite via bun:sqlite.

Configuration

See src/configuration.ts for all options.

Required

  • CLI_API_TOKEN - Base shared secret used by CLI, web login and native pairing. Clients append :<namespace> for isolation. Auto-generated on first run if not set.

Optional (Telegram)

  • TELEGRAM_BOT_TOKEN - Token from @BotFather.
  • HAPI_PUBLIC_URL - Public HTTPS URL for Telegram Mini App access. Also used to derive default CORS origins for the web app.

Optional (Voice)

Dictation and voice-assistant provider keys can also be added from Settings → Voice (stored in settings.json under providerCredentials; env vars still win when set at process start).

  • ELEVENLABS_API_KEY - ElevenLabs API key for voice assistant + dictation.
  • ELEVENLABS_AGENT_ID - Custom ElevenLabs agent ID (auto-created if not set).
  • GEMINI_API_KEY / GOOGLE_API_KEY - Gemini Live voice assistant.
  • DASHSCOPE_API_KEY / QWEN_API_KEY - Qwen Realtime voice assistant.
  • VOICE_BACKEND - Default assistant backend (elevenlabs, gemini-live, or qwen-realtime).
  • OPENAI_API_KEY - OpenAI dictation (gpt-transcribe / gpt-live-transcribe).
  • DEEPGRAM_API_KEY - Deepgram dictation (nova-3, standard and realtime).
  • GROQ_API_KEY - Groq dictation (whisper-large-v3).
  • TRANSCRIPTION_BASE_URL and TRANSCRIPTION_MODEL - OpenAI-compatible/local transcription endpoint and model.
  • TRANSCRIPTION_API_KEY - Optional bearer token for the OpenAI-compatible endpoint.

Optional

  • HAPI_LISTEN_HOST - HTTP bind address (default: 127.0.0.1).
  • HAPI_LISTEN_PORT - HTTP port (default: 3006).
  • CORS_ORIGINS - Comma-separated origins, or *.
  • HAPI_HOME - Data directory (default: ~/.hapi).
  • DB_PATH - SQLite database path (default: HAPI_HOME/hapi.db).
  • TELEGRAM_NOTIFICATION - Enable/disable Telegram notifications (default: true).
  • HAPI_RELAY_API - Relay API domain (default: relay.hapi.run).
  • HAPI_RELAY_AUTH - Explicit relay auth key. By default the hub obtains and persists an individually revocable key from the relay. A persisted key rejected with HTTP 403 is discarded and reissued once; an explicitly configured environment key must be updated manually.
  • HAPI_RELAY_FORCE_TCP - Force TCP relay mode (true/1).
  • VAPID_SUBJECT - Contact email/URL for Web Push.
  • HAPI_ANDROID_PUSH - auto (default: direct FCM when credentials are configured, otherwise relay), relay, fcm, or off.
  • FCM_SERVICE_ACCOUNT_PATH - Service-account JSON for private Firebase builds; the app must use the same project. Invalid configured credentials disable Android push instead of switching projects.
  • HAPI_IOS_PUSH - relay (default), apns, or off.
  • HAPI_PUSH_RELAY_URL - Shared Android/iOS push relay (default: https://push.hapi.run; persisted as iosPushRelayUrl). Independent of the --relay network tunnel.
  • HAPI_SESSION_IDLE_TIMEOUT_MS - Keep-alive-idle window in ms (default: 43200000 / 12 h; 0 disables). See "Session liveness" below.

Official native apps register their encryption keys automatically; no push provider setup is needed on a fresh hub. See the native companion push contract.

Running

Binary (single executable):

export TELEGRAM_BOT_TOKEN="..."
export CLI_API_TOKEN="shared-secret"
export HAPI_PUBLIC_URL="https://your-domain.example"

hapi hub

hapi server remains supported as an alias.

For web/native clients + CLI, you can omit TELEGRAM_BOT_TOKEN. To enable Telegram, set TELEGRAM_BOT_TOKEN and HAPI_PUBLIC_URL, start the hub, open /app in the bot chat, and bind the Mini App with CLI_API_TOKEN:<namespace> when prompted.

From source:

bun install
bun run dev:hub

HTTP API

The following is an overview. See the client contract for request/response shapes and error semantics, and src/web/routes/ for all endpoints.

Authentication (src/web/routes/auth.ts, src/web/routes/bind.ts)

  • POST /api/auth - Get JWT token (Telegram initData or CLI_API_TOKEN[:namespace]).
  • POST /api/bind - Bind a Telegram account using initData + CLI_API_TOKEN:<namespace>.

Sessions (src/web/routes/sessions.ts)

  • GET /api/sessions - List all sessions. Each summary includes hasConversationContent, derived from stored conversation messages (not titles or lifecycle events); full session SSE updates carry changes to this flag.
  • GET /api/sessions/:id - Get session details.
  • POST /api/sessions/:id/abort - Abort session.
  • POST /api/sessions/:id/switch - Hand off session control to the web.
  • POST /api/sessions/:id/resume - Resume inactive session.
  • POST /api/sessions/:id/reopen - Reopen a session; follow the returned session ID.
  • POST /api/sessions/:id/clear - Start a fresh conversation when supported.
  • POST /api/sessions/:id/upload - Upload file (base64, max 50MB).
  • POST /api/sessions/:id/upload/delete - Delete uploaded file.
  • POST /api/sessions/:id/archive - Archive active session.
  • PATCH /api/sessions/:id - Rename session.
  • DELETE /api/sessions/:id - Delete inactive session.
  • GET /api/sessions/:id/slash-commands - List slash commands.
  • GET /api/sessions/:id/skills - List skills.
  • POST /api/sessions/:id/permission-mode - Set permission mode.
  • POST /api/sessions/:id/model - Set model preference.
  • POST /api/sessions/:id/effort - Set effort for Claude, Grok, or Pi; other config controls are flavor/capability-gated (see the client contract).
  • GET/POST /api/sessions/:id/scratchlist - Read/create Hub-persisted scratchlist entries; update/delete and attachment routes share this prefix.

Messages (src/web/routes/messages.ts)

  • GET /api/sessions/:id/messages - Get messages (paginated).
  • POST /api/sessions/:id/messages - Send message.
  • POST /api/sessions/:id/messages/queued-state - Reconcile queued, indeterminate, and invoked local IDs.
  • DELETE /api/sessions/:id/messages/:messageId - Cancel a queued message.
  • POST /api/sessions/:id/messages/:messageId/steer - Steer a queued message when the session supports it.
  • POST /api/sessions/:id/messages/:messageId/retry - Explicitly retry indeterminate delivery when safe; never auto-replay.

Permissions (src/web/routes/permissions.ts)

  • POST /api/sessions/:id/permissions/:requestId/approve - Approve permission.
  • POST /api/sessions/:id/permissions/:requestId/deny - Deny permission.

Machines (src/web/routes/machines.ts)

  • GET /api/machines - List online machines.
  • PATCH /api/machines/:id - Set/clear the machine display name.
  • GET /api/machines/:id/agent-availability - List installed/configured Agents.
  • POST /api/machines/:id/spawn - Spawn new session on machine.
  • POST /api/machines/:id/list-directory - Browse runner-scoped directories.
  • POST /api/machines/:id/paths/exists - Check if path exists.
  • POST /api/machines/:id/restart-runner - Request runner restart (requires an available restart path).

Usage (src/web/routes/usage.ts)

  • GET /api/usage/summary - Get cache-aware token usage for the owner namespace (range=7d|30d|all).

Storage (src/web/routes/storage.ts)

  • GET /api/storage/sqlite - SQLite database/WAL/SHM sizes for the owner namespace.

Git/Files (src/web/routes/git.ts)

  • GET /api/sessions/:id/git-status - Git status.
  • GET /api/sessions/:id/git-diff-numstat - Diff summary.
  • GET /api/sessions/:id/git-diff-file - File-specific diff.
  • GET /api/sessions/:id/file - Read file content.
  • GET /api/sessions/:id/files - File search with ripgrep.

Events (src/web/routes/events.ts)

  • GET /api/events - SSE stream for live updates.
  • POST /api/visibility - Report client visibility state.

Voice (src/web/routes/voice.ts)

  • POST /api/voice/token - Get ElevenLabs conversation token.
  • GET /api/voice/backend - Discover configured assistant backends.
  • GET /api/voice/voices - List voices for the selected backend.
  • GET/PUT /api/voice/transcription/credentials - Read masked provider settings or update credentials (owner-only).
  • GET /api/voice/transcription/providers - List configured providers and supported modes.
  • POST /api/voice/transcription - Transcribe a bounded recording.
  • POST /api/voice/transcription/realtime-token - Mint a short-lived OpenAI, ElevenLabs, or Deepgram credential.

Push Notifications (src/web/routes/push.ts)

  • GET /api/push/vapid-public-key - Get VAPID public key.
  • POST /api/push/subscribe - Subscribe to push notifications.
  • DELETE /api/push/subscribe - Unsubscribe.

Native Android/iOS registration uses POST/DELETE /api/devices/register (src/web/routes/devices.ts); see the native push contract.

CLI (src/web/routes/cli.ts)

  • POST /cli/sessions - Create/load session.
  • GET /cli/sessions/:id - Get session by ID.
  • POST /cli/machines - Create/load machine.
  • GET /cli/machines/:id - Get machine by ID.

Socket.IO

See src/socket/handlers/cli/index.ts and src/socket/handlers/terminal.ts for event handlers.

Namespaces: /cli (raw CLI access token) and /terminal (client JWT). Ordinary web/native session updates use SSE, not the CLI Socket.IO namespace.

Client events (CLI to hub)

  • message - Send message to session.
  • update-metadata - Update session metadata.
  • update-state - Update agent state.
  • session-alive - Keep session active.
  • session-ready - Cursor ACP session/load (or newSession) succeeded; hub defers merge/dedup until this arrives on reopen.
  • session-end - Mark session ended.
  • machine-alive - Keep machine online.
  • rpc-register - Register RPC handler.
  • rpc-unregister - Unregister RPC handler.

Terminal events (web to hub, /terminal)

  • terminal:create - Open terminal for session.
  • terminal:write - Send input.
  • terminal:resize - Resize dimensions.
  • terminal:close - Close terminal.
  • agent-terminal:subscribe / agent-terminal:unsubscribe - Attach/detach the wrapped agent's terminal stream.
  • agent-terminal:input / agent-terminal:resize - Interact with that terminal when supported.

Hub events (hub to CLI clients, /cli)

  • update - Broadcast session/message updates.
  • rpc-request - Incoming RPC call.

See src/socket/rpcRegistry.ts for RPC routing.

Telegram Bot

See src/telegram/bot.ts for bot implementation.

Commands

  • /start - Welcome message with Mini App link.
  • /app - Open Mini App.

Features

  • Permission request notifications with approve/deny buttons.
  • Session ready notifications.
  • Deep links to Mini App sessions.

See src/telegram/callbacks.ts for button handlers.

Core Logic

See src/sync/syncEngine.ts for the main session/message manager:

  • In-memory session cache with versioning.
  • Message pagination and retrieval.
  • Permission approval/denial.
  • RPC method routing via Socket.IO.
  • Event publishing to SSE and Telegram.
  • Git operations and file search.
  • Activity tracking and timeouts.

Session liveness

Two separate signals, deliberately (tiann/hapi#1820):

  • active / activeAt - transport. The CLI socket is connected; session-alive refreshes it every ~2 s. sessionCache.expireInactive drops active after 30 s of silence.
  • metadata.lifecycleState - agent health. running -> idle -> running, plus archived.

Keep-alives prove a socket, not an agent. A session can therefore heartbeat for days with no messages, no thinking and no background tasks - active: true forever, which expireInactive can never catch. sessionCache.reconcileKeepaliveIdle (same 5 s tick) closes that gap: after HAPI_SESSION_IDLE_TIMEOUT_MS with no agent progress it moves lifecycleState running -> idle.

  • active is left alone on purpose. The CLI really is reachable, and flipping it would let resumeSession spawn a second agent against a live process, and unlock dedup-merge / delete on it.
  • Progress means a message in either direction, a queued prompt, or a background task - never a keep-alive. Sessions that are thinking, running background tasks or holding a pending permission request are never marked.
  • idle reverts to running on the next progress, within one tick.
  • metadata.idleReconcileExempt: true opts a session out entirely.

Storage

See src/store/index.ts for SQLite persistence:

  • Sessions with metadata and agent state.
  • Messages with pagination support.
  • Machines with runner state.
  • Todo extraction from messages.
  • Users table for Telegram bindings (includes namespace).
  • Scratchlist entries/attachments, usage, work graph, and push registrations.

Message content is stored via src/store/contentCodec.ts: oversized strings inside agent messages (giant tool output) are head+tail truncated at ingest, and payloads ≥256 chars are zstd-compressed (TEXT = plaintext JSON, BLOB = zstd). User prompts are never truncated — queued rows are delivered to the CLI verbatim.

Maintenance scripts (run with the hub stopped before swapping files):

  • scripts/compact-db.ts — retroactively truncate + compress + VACUUM an existing DB into a new file (the source is only opened read-only).
  • scripts/cleanup-sessions.ts — bulk-delete sessions by message count, path glob, or first-message pattern.

Source structure

  • src/web/ - HTTP service and routes.
  • src/socket/ - Socket.IO setup and handlers.
  • src/socket/handlers/cli/ - Modular CLI handlers.
  • src/telegram/ - Telegram bot.
  • src/sync/ - Core session/message logic.
  • src/store/ - SQLite persistence.
  • src/sse/ - Server-Sent Events.
  • src/config/ - Configuration loading and generation.
  • src/notifications/ - Push and Telegram notifications.
  • src/visibility/ - Client visibility tracking.

Security model

Access is controlled by:

  • Telegram initData verification plus bound Telegram users (bound via CLI_API_TOKEN:<namespace>).
  • CLI_API_TOKEN base secret for CLI and browser access (namespace is appended by clients).

Transport security depends on HTTPS in front of the hub.

Build for deployment

From the repo root:

bun run build:hub
bun run build:web

The hub build output is hub/dist/index.js, and the web assets are in web/dist.

Networking notes

  • Telegram Mini Apps require HTTPS and a public URL. If the hub has no public IP, use Cloudflare Tunnel or Tailscale and set HAPI_PUBLIC_URL to the HTTPS endpoint.
  • If the web app is hosted on a different origin, set CORS_ORIGINS (or HAPI_PUBLIC_URL) to include that static host origin.

Standalone web hosting

The web UI can be hosted separately from the hub (for example on GitHub Pages or Cloudflare Pages):

  1. Build and deploy web/dist from the repo root.
  2. Set CORS_ORIGINS (or HAPI_PUBLIC_URL) to the static host origin.
  3. Open the static site, click the Hub button on the login screen, and enter the hapi hub origin.

Leaving the hub override empty preserves the default same-origin behavior when the hub serves the web assets directly.