How this roadmap works
Shipped = verified by the live changelog and the probe harness. Rolling out = work-in-progress with a target quarter. Future = direction, not commitment. Dates slip; priorities reorder. The system itself ships 1-6 improvements every day via the eternal-loop, so this evolves continuously.
Today (verified live)
Discord chat
Persistent bot, per-user memory, anti-fab post-process.
Email integration
Gmail OAuth read + send. Triage handler.
Headless browser autopilot
Playwright + Chromium. Goal-driven nav, click, fill, extract.
Code-execution sandbox
Ephemeral Docker container with pandas/numpy. Network-isolated, memory-capped.
Persistent memory + vault
Tessarion cross-platform vault + local conversation archive.
244 specialist agents
Domain specialists you can delegate to. Parallel missions.
Eternal-improvement loop
Auto-ships code upgrades every 30 min. Watchdog rollback.
Anti-fabrication (3 tiers)
Catches action claims that aren't backed by real tool calls.
8 proactive monitors
Cost spike, error burst, daily summary, eternal-loop alerts, more.
Pro Max routing (zero marginal cost)
Cloud-local Claude Code subprocess via OAuth token.
Q3 2026 (in progress)
SMS + voice (Twilio)
Real outbound SMS for monitor alerts. Inbound webhook already wired. ~$15/mo for a phone number.
Cron scheduler tool
"Remind me at 3pm" / "every morning at 8 summarize my inbox." Fire-and-forget for future-tense work.
Background task queue
"Research X for an hour, ping me when done." Long-running jobs survive chat closing.
Image input in chat
Drop a screenshot in Discord, Stratam sees it. Multi-modal understanding for ambiguous prompts.
Multi-tenant + Stripe billing
User accounts, per-user isolation, subscription billing. Needed before paid tiers open.
Onboarding wizard
Multi-step setup: paste Discord bot token, pick timezone, connect Gmail. Replaces .env editing.
Q4 2026 — Q1 2027
Modular refactor
Break the monolith into focused modules. Easier testing, community contributions, plugin architecture.
Computer-use agent
Real desktop via Xvfb + Anthropic's Computer Use API. For sites that resist Playwright DOM automation.
Phone-call mode (Twilio Voice + Whisper)
Stratam takes/makes real phone calls. Whisper transcribes, ElevenLabs speaks. Voicemail summarization.
Mobile bridge (iOS/Android)
ADB on Android, Shortcuts on iOS. Real phone control. Big "do anything" unlock for non-web tasks.
Banking + payments (Plaid)
Real account access for budgeting, subscription tracking, transaction categorization. High trust bar required.
Long-running autonomous projects
Give Stratam a goal, he works for days, reports back with revertable changes. Closest thing to "an actual employee."
What we're NOT building
Some things people ask about that we've decided against:
- Our own foundation model. We route to Claude / GPT / Gemini and stack on top. Building a competing LLM is a $100M+ bet — not the right shape for a 1-founder team.
- Native mobile app for Stratam itself. The web phone view + Discord/SMS/voice channels cover this. A dedicated iOS/Android app adds maintenance overhead for low marginal value.
- Generic "AI for X" verticals. Stratam is general-purpose by design. Vertical wrappers (AI for sales, AI for legal, etc.) compete on prompt engineering, not the operational substrate.
- An AI gold-rush feature for every new model release. We add capabilities when there's a real operator gap, not because OpenAI shipped o3 yesterday.