Voice memos. Natural language commands. Real-time alerts. Agent dispatch. All from Slack. All from your phone.
Your 11 AI agents are always listening. Just tell them what to do.
Get Voice for Slack →Included in Lighthouse ($2.5K/mo) and Fleet ($6.5K/mo)
Live Client
Ziv Health manages 11 AI agents from Slack
Telehealth platform. 139 cron jobs. 150+ event types. One operator. One Slack channel.
Record a voice memo or type a message in your #ops channel. @mention the bot or DM it directly.
Two-tier NLP: regex patterns for fast match, Claude Haiku for ambiguous queries. Intent mapped to a registered soul action.
The target soul runs the action — query data, pause campaigns, check compliance. Result posted back in the same Slack thread.
Voice for Slack replaces your dashboard. Every metric, every alert, every agent command — accessible via natural language in Slack.
Type or speak. Your agents understand both.
Two-tier intent classification: pattern matching for common queries (<10ms, zero cost), Claude Haiku fallback for ambiguous requests (~1s, $0.003). 15+ intents recognized automatically.
Record a voice note. Agents read the transcript and act.
Walk, commute, or move between meetings. Record a voice memo in Slack. Shulam transcribes it via Whisper, classifies the intent, dispatches to the right agent, and posts the result back in-thread.
10+ registered actions. Destructive actions require confirmation.
Each soul registers safe actions it can execute on command. 'Pause ads' requires double-confirmation within 5 minutes. 'Show MRR' executes instantly. Every action is logged in the HIPAA audit trail.
Payments, churn, disputes, and refills — all in your Slack channel.
Every payment posts within 5 seconds. Churn risk alerts include the patient's negative sampling cohort and estimated MRR loss. Stripe disputes trigger critical alerts with a 7-day deadline. No dashboard needed — Slack is your command center.
G-Stack scorecard every morning at 7am ET.
Revenue WoW, retention at 30d and 90d, CAC, Default Alive status, fleet health — all in one formatted Slack message. Red metrics get flagged. Green metrics get confirmed. One scroll covers your entire business.
Pause campaigns, process refunds — only with confirmation.
Destructive actions (pause ads, cancel subscriptions, issue refunds) require explicit confirmation within a 5-minute window. The operator types 'confirm' to proceed. No accidental actions. User Sovereignty enforced.
Just type or speak. No slash commands. No syntax. Plain English.
"What's the MRR?"
→ Revenue Analyst
MRR: $12,400 | WoW: +7.2% | 43 active subs
"Any at-risk patients?"
→ Retention Agent
3 at risk, 7 refills due. Win-back triggered.
"Pause the ad budget"
→ Ad Optimizer
⚠️ Confirmation required. Reply "confirm" within 5 min.
"Compliance status"
→ Compliance Monitor
HIPAA: auditing. FDA: scanning. LegitScript: approved.
"How's the fleet?"
→ System
11/11 souls reporting. 0 degraded. 0 down.
"Pharmacy fulfillment?"
→ Inventory Watcher
MediVera: 25 offerings active. Webhook verified.
Most operators manage their AI agents through dashboards. Dashboards require a laptop, a browser, and uninterrupted attention. Voice for Slack removes all three requirements. You manage your entire agent fleet from your phone, while walking, commuting, or between meetings. The shift from dashboard to voice is the same shift that happened when mobile banking replaced branch visits — the capability is identical, but the friction drops to zero.
The technical foundation is a two-tier natural language processing pipeline. Tier 1 uses regex pattern matching against 15 intent categories — this handles 80 percent of queries in under 10 milliseconds at zero cost. Tier 2 uses Claude Haiku for ambiguous or complex queries, adding approximately one second of latency at a cost of $0.003 per classification. The combination delivers high accuracy without the latency or cost of routing every message through a large language model.
Voice memos are transcribed using Whisper, the same speech recognition model used by leading transcription services. The transcript is then processed through the same two-tier classification pipeline as text messages. The result is that voice and text are functionally equivalent — the operator can switch between them without changing behavior or losing capability. This is important because voice is faster for some contexts (walking, driving) and text is faster for others (meetings, quiet environments).
The agent dispatch layer is the key architectural innovation. Each soul registers the actions it can execute — query actions (read-only) and command actions (state-changing). Command actions that are destructive (pausing ad campaigns, processing refunds, cancelling subscriptions) require explicit confirmation within a five-minute window. This implements the G-Stack principle of User Sovereignty: AI recommends and executes, but the operator always makes the final decision on irreversible actions. The confirmation window prevents accidental commands while maintaining the speed advantage of voice-based operation.
For Shulam clients, Voice for Slack is the command interface that scales with the agent fleet. When a new soul is deployed, it registers its actions with the dispatch service, and those actions become immediately available via voice and text in Slack. No UI changes, no dashboard updates, no deployment required. The operator simply asks the new soul what it can do, and it responds. This is how autonomous operations should work — the interface adapts to the agents, not the other way around.
Voice for Slack is included in all Shulam AaaS tiers. No additional setup. No additional cost.