Remove freebuff waiting room flag

.env.example

@@ -32,6 +32,9 @@ LINKUP_API_KEY=dummy_linkup_key
 LOOPS_API_KEY=dummy_loops_key
 ZEROCLICK_API_KEY=dummy_zeroclick_key
 
+# Freebuff
+FREEBUFF_SESSION_LENGTH_MS=3600000
+
 # Discord Integration
 DISCORD_PUBLIC_KEY=dummy_discord_public_key
 DISCORD_BOT_TOKEN=dummy_discord_bot_token

cli/src/app.tsx

@@ -261,7 +261,7 @@ export const App = ({
   }
 
   // Render project picker FIRST when at home directory or outside a project.
-  // This deliberately precedes the login/auth and waiting-room gates so the
+  // This deliberately precedes the login/auth and freebuff session gate so the
   // user always gets to pick a working directory before anything else — auth
   // failures or a banned/queued freebuff session would otherwise replace the
   // picker mid-flash and look like being kicked out of the app.
@@ -340,7 +340,7 @@ interface AuthedSurfaceProps {
 }
 
 /**
- * Rendered only after auth is confirmed. Owns the freebuff waiting-room gate
+ * Rendered only after auth is confirmed. Owns the freebuff session gate
  * so `useFreebuffSession` runs exactly once per authed session (not before
  * we have a token).
  */

cli/src/components/waiting-room-screen.tsx

@@ -496,11 +496,11 @@ export const WaitingRoomScreen: React.FC<WaitingRoomScreenProps> = ({
             </box>
           )}
 
-          {/* Server says the waiting room is disabled — this screen should not
-              normally render in that case, but show a minimal message just in
-              case App.tsx's guard is bypassed. */}
+          {/* Compatibility fallback for older servers without the session
+              endpoint. This should not normally render because App.tsx treats
+              it as admitted. */}
           {session?.status === 'disabled' && (
-            <text style={{ fg: theme.muted }}>Waiting room disabled.</text>
+            <text style={{ fg: theme.muted }}>Session gate unavailable.</text>
           )}
 
           {/* Country outside the free-mode allowlist. Terminal — polling has

cli/src/hooks/helpers/send-message.ts

@@ -523,7 +523,7 @@ export const handleRunError = (params: {
 }
 
 /**
- * Surface + recover from a waiting-room gate rejection. The server rejected
+ * Surface + recover from a freebuff session gate rejection. The server rejected
  * the request because our seat is no longer valid; update local state so the
  * UI reflects reality and we stop sending requests until we re-admit.
  */

cli/src/hooks/use-freebuff-session.ts

@@ -73,8 +73,7 @@ async function callSession(
     signal: opts.signal,
   })
   // 404 = endpoint not deployed on this server (older web build). Treat as
-  // "waiting room disabled" so a newer CLI against an older server still
-  // works, rather than stranding users in a waiting room forever.
+  // a compatibility bypass so a newer CLI against an older server still works.
   if (resp.status === 404) {
     return { status: 'disabled' }
   }

cli/src/utils/error-handling.ts

@@ -96,7 +96,7 @@ export const getCountryBlockFromFreeModeError = (
 }
 
 /**
- * Freebuff waiting-room gate errors returned by /api/v1/chat/completions.
+ * Freebuff session gate errors returned by /api/v1/chat/completions.
  *
  * Contract (see docs/freebuff-waiting-room.md):
  *   - 428 `waiting_room_required`   — no session row exists; POST /session to join.

common/src/types/freebuff-session.ts

@@ -1,11 +1,9 @@
 import type { FreebuffAccessTier } from '../constants/freebuff-models'
 
 /**
- * Wire-level shapes returned by `/api/v1/freebuff/session`. Source of truth
- * for the CLI (which deserializes these) and the server (which serializes
- * them) — keep both in sync by importing this module from either side.
- *
- * The CLI uses these shapes directly; there are no client-only states.
+ * Shapes used by `/api/v1/freebuff/session` and the CLI. Most variants are
+ * wire-level responses serialized by the server; explicitly documented
+ * compatibility variants may be synthesized by the CLI for older servers.
  */
 
 /**
@@ -67,8 +65,8 @@ export type FreebuffIpPrivacySignal =
 
 export type FreebuffSessionServerResponse =
   | {
-      /** Waiting room is globally off; free-mode requests flow through
-       *  unchanged. Client should treat this as "admitted forever". */
+      /** Compatibility fallback for older servers without the session
+       *  endpoint. Client should treat this as "admitted forever". */
       status: 'disabled'
     }
   | {

docs/freebuff-waiting-room.md

@@ -8,22 +8,14 @@ The waiting room is the admission control layer for **free-mode** requests again
 2. **Gate on per-deployment health and hours** — a single fleet probe per tick (`getFleetHealth` in `web/src/server/free-session/fireworks-health.ts`) hits the Fireworks metrics endpoint and classifies each dedicated deployment as `healthy | degraded | unhealthy`. Only models whose deployment is `healthy` and currently available admit that tick; GLM 5.1 is available during 9am ET-5pm PT on weekdays, while MiniMax M2.7 is serverless and always available.
 3. **One instance per account** — prevent a single user from running N concurrent freebuff CLIs to get N× throughput.
 
-Users who cannot be admitted immediately are placed in the queue for their chosen model and given an estimated wait time. Admitted users get a fixed-length session (default 1h) bound to the model they were admitted on; chat completions use that model for the life of the session.
+Users who cannot be admitted immediately are placed in the queue for their chosen model and given an estimated wait time. With the current high instant-admit capacities, most users go straight from model selection to an active session; the queue only appears when a model is actually saturated. Admitted users get a fixed-length session (default 1h) bound to the model they were admitted on; chat completions use that model for the life of the session.
 
-The entire system is gated by the env flag `FREEBUFF_WAITING_ROOM_ENABLED`. When `false`, the gate is a no-op and the admission ticker does not start; free-mode traffic flows through unchanged.
-
-## Kill Switch
+## Configuration
 
 ```bash
-# Disable entirely (both the gate on chat/completions and the admission loop)
-FREEBUFF_WAITING_ROOM_ENABLED=false
-
-# Other knob (only read when enabled)
 FREEBUFF_SESSION_LENGTH_MS=3600000         # 1 hour
 ```
 
-Flipping the flag is safe at runtime: existing rows stay in the DB and will be admitted / expired correctly whenever the flag is flipped back on.
-
 ## Architecture
 
 ```mermaid
@@ -186,9 +178,6 @@ Before any of those state transitions, the handler requires a resolved allowlist
 Response shapes:
 
 ```jsonc
-// Waiting room disabled — CLI should treat this as "always admitted"
-{ "status": "disabled" }
-
 // In queue
 {
   "status": "queued",
@@ -272,9 +261,7 @@ For free-mode requests (`codebuff_metadata.cost_mode === 'free'`), `_post.ts` ca
 | 409  | `session_superseded`       | Claimed `instance_id` does not match stored one — another CLI took over.                                                                       |
 | 410  | `session_expired`          | `expires_at + grace < now()` (past the hard cutoff). Client should POST /session to re-queue.                                                  |
 
-Successful results carry one of three reasons: `disabled` (gate is off), `active` (`expires_at > now()`, `remainingMs` provided), or `draining` (`expires_at <= now() < expires_at + grace`, `gracePeriodRemainingMs` provided). The CLI should treat `draining` as "let any in-flight agent run finish, but block new user prompts" — see [Drain / Grace Window](#drain--grace-window) below. The corresponding wire status from `getSessionState` is `ended`.
-
-When the waiting room is disabled, the gate returns `{ ok: true, reason: 'disabled' }` without touching the DB.
+Successful results carry one of two reasons: `active` (`expires_at > now()`, `remainingMs` provided), or `draining` (`expires_at <= now() < expires_at + grace`, `gracePeriodRemainingMs` provided). The CLI should treat `draining` as "let any in-flight agent run finish, but block new user prompts" — see [Drain / Grace Window](#drain--grace-window) below. The corresponding wire status from `getSessionState` is `ended`.
 
 ## Drain / Grace Window
 
@@ -314,8 +301,6 @@ The CLI:
 8. **Handles chat-gate errors:** the same statuses are reachable via the gate's 409/410/428/429 for fast in-flight feedback, and the CLI calls the matching `markFreebuff*` helper to flip local state without waiting for the next poll.
 9. **On clean exit**, calls `DELETE /api/v1/freebuff/session` so the next user can be admitted sooner.
 
-The `disabled` response means the server has the waiting room turned off. CLI treats it identically to `active` with infinite remaining time — no countdown, and chat requests can omit `freebuff_instance_id` entirely.
-
 ## Multi-pod Behavior
 
 - **`/api/v1/freebuff/session` routes** are stateless per pod; all state lives in Postgres. Any pod can serve any request.

packages/internal/src/env-schema.ts

@@ -56,14 +56,6 @@ export const serverEnvSchema = clientEnvSchema.extend({
   // sweep but risks rate-limiting.
   BOT_SWEEP_GITHUB_TOKEN: z.string().min(1).optional(),
 
-  // Freebuff waiting room. Defaults to OFF so the feature requires explicit
-  // opt-in per environment — the CLI/SDK do not yet send
-  // freebuff_instance_id, so enabling this before they ship would reject
-  // every free-mode request with 428 waiting_room_required.
-  FREEBUFF_WAITING_ROOM_ENABLED: z
-    .enum(['true', 'false'])
-    .default('false')
-    .transform((v) => v === 'true'),
   FREEBUFF_SESSION_LENGTH_MS: z.coerce
     .number()
     .int()
@@ -136,8 +128,7 @@ export const serverProcessEnv: ServerInput = {
   BOT_SWEEP_SECRET: process.env.BOT_SWEEP_SECRET,
   BOT_SWEEP_GITHUB_TOKEN: process.env.BOT_SWEEP_GITHUB_TOKEN,
 
-  // Freebuff waiting room
-  FREEBUFF_WAITING_ROOM_ENABLED: process.env.FREEBUFF_WAITING_ROOM_ENABLED,
+  // Freebuff session gate
   FREEBUFF_SESSION_LENGTH_MS: process.env.FREEBUFF_SESSION_LENGTH_MS,
   FREEBUFF_DEV_FORCE_LIMITED: process.env.FREEBUFF_DEV_FORCE_LIMITED,
 }

web/src/app/api/v1/chat/completions/__tests__/completions.test.ts

@@ -82,11 +82,10 @@ describe('/api/v1/chat/completions POST endpoint', () => {
   let mockInsertMessageBigquery: InsertMessageBigqueryFn
   let nextQuotaReset: string
 
-  // Bypasses the freebuff waiting-room gate in tests that exercise free-mode
-  // flow without seeding a session. Matches the real return for the disabled
-  // path so downstream logic proceeds normally.
+  // Bypasses the freebuff session gate in tests that exercise free-mode flow
+  // without seeding a session.
   const mockCheckSessionAdmissibleAllow = async () =>
-    ({ ok: true, reason: 'disabled' }) as const
+    ({ ok: true, reason: 'active', remainingMs: 60 * 60 * 1000 }) as const
   const mockResolveFreeModeCountryAccess = async (
     _userId: string,
     req: Parameters<typeof getFreeModeCountryAccess>[0],

web/src/app/api/v1/chat/completions/_post.ts

@@ -183,7 +183,7 @@ export async function postChatCompletions(params: {
     logger: Logger
   }) => Promise<BlockGrantResult | null>
   getUserPreferences?: GetUserPreferencesFn
-  /** Optional override for the freebuff waiting-room gate. Defaults to the
+  /** Optional override for the freebuff session gate. Defaults to the
    *  real check backed by Postgres; tests inject a no-op. */
   checkSessionAdmissible?: CheckSessionAdmissibleFn
   /** Optional override for the free-mode rate limiter. Tests inject this to
@@ -527,16 +527,14 @@ export async function postChatCompletions(params: {
 
     let freeModeSessionGate: SessionGateResult | null = null
 
-    // Freebuff waiting-room gate. Usually enforced only when
-    // FREEBUFF_WAITING_ROOM_ENABLED=true. Runs before the rate limiter so
-    // rejected requests don't burn a queued user's free-mode counters.
+    // Freebuff session gate. Runs before the rate limiter so rejected requests
+    // don't burn a queued user's free-mode counters.
     if (isFreeModeRequest) {
       const claimedInstanceId =
         typedBody.codebuff_metadata?.freebuff_instance_id
       freeModeSessionGate = await checkSession({
         userId,
         accessTier: freebuffAccessTier,
-        userEmail: userInfo.email,
         claimedInstanceId,
         requestedModel: typedBody.model,
         requireActiveSession: isFreebuffGeminiThinkerAgent(agentId),

web/src/app/api/v1/freebuff/session/__tests__/session.test.ts

@@ -102,7 +102,6 @@ function makeSessionDeps(overrides: Partial<SessionDeps> = {}): SessionDeps & {
   let instanceCounter = 0
   return {
     rows,
-    isWaitingRoomEnabled: () => true,
     graceMs: 30 * 60 * 1000,
     sessionLengthMs: 60 * 60 * 1000,
     // Keep instant-admit disabled in handler tests — they verify queue/state
@@ -225,16 +224,6 @@ describe('POST /api/v1/freebuff/session', () => {
     })
   })
 
-  test('returns disabled when waiting room flag is off', async () => {
-    const sessionDeps = makeSessionDeps({ isWaitingRoomEnabled: () => false })
-    const resp = await postFreebuffSession(
-      makeReq('ok'),
-      makeDeps(sessionDeps, 'u1'),
-    )
-    const body = await resp.json()
-    expect(body.status).toBe('disabled')
-  })
-
   test('creates a limited DeepSeek Flash session for disallowed country', async () => {
     const sessionDeps = makeSessionDeps()
     const resp = await postFreebuffSession(

web/src/app/api/v1/freebuff/session/_handlers.ts

@@ -74,7 +74,7 @@ export interface FreebuffSessionDeps {
 
 type AuthResult =
   | { error: NextResponse }
-  | { userId: string; userEmail: string | null; userBanned: boolean }
+  | { userId: string; userBanned: boolean }
 
 async function resolveUser(
   req: NextRequest,
@@ -94,7 +94,7 @@ async function resolveUser(
   }
   const userInfo = await deps.getUserInfoFromApiKey({
     apiKey,
-    fields: ['id', 'email', 'banned'],
+    fields: ['id', 'banned'],
     logger: deps.logger,
   })
   if (!userInfo?.id) {
@@ -107,7 +107,6 @@ async function resolveUser(
   }
   return {
     userId: String(userInfo.id),
-    userEmail: userInfo.email ?? null,
     userBanned: Boolean(userInfo.banned),
   }
 }
@@ -160,7 +159,6 @@ export async function postFreebuffSession(
   try {
     const state = await requestSession({
       userId: auth.userId,
-      userEmail: auth.userEmail,
       userBanned: auth.userBanned,
       model: requestedModel,
       accessTier,
@@ -207,7 +205,6 @@ export async function getFreebuffSession(
     const state = await getSessionState({
       userId: auth.userId,
       accessTier,
-      userEmail: auth.userEmail,
       userBanned: auth.userBanned,
       claimedInstanceId,
       deps: deps.sessionDeps,
@@ -244,7 +241,6 @@ export async function deleteFreebuffSession(
   try {
     await endUserSession({
       userId: auth.userId,
-      userEmail: auth.userEmail,
       deps: deps.sessionDeps,
     })
     return NextResponse.json({ status: 'ended' }, { status: 200 })