Best practices
After watching users wire Stratam up a thousand ways, here's what tends to work.
Seed the vault early
The single highest-leverage thing you can do in week one: tell Stratam who you are and what you're working on. Drop notes like:
- "I'm a solo founder building a B2B SaaS for HVAC contractors."
- "My pricing is $99/mo per technician, billed annually."
- "Our key competitor is ServiceTitan but we focus on the 1-10 employee segment."
- "I'm bullish on RWAs and skeptical of memecoins."
When Stratam answers anything related, it calls recall first and grounds the reply in what you've shared. Without context, you get generic AI. With context, you get an assistant that talks like it works for you.
Treat monitors like cron + an agent
A monitor is a scheduled prompt + a notify channel. The two failure modes:
Too vague → "Check the news" returns generic news every hour. Useless.
Too specific → "Visit www.foo.com/page-3, click the second button, then extract the price" breaks the first time the page changes.
The sweet spot: a clear goal + flexibility in how the agent gets there.
"Check BTC/USD via web_search. Compare to your last check (recall 'last BTC price'). If it moved more than 3% in either direction, reply with the % change and the dollar value. Otherwise reply 'no significant move' and save the new price via remember."
Pin what matters
/conversations lets you pin threads to the top. Use it for:
- Long-running projects (a customer's onboarding, a launch you're planning)
- Important reference threads (the one where you decided your pricing)
- Active investigations (debugging a production issue)
Pinned threads survive scroll. Unpinned threads fall off the recent list after enough activity elsewhere.
Use the right channel for the message
Stratam adapts brevity to the channel automatically:
- SMS: 1-3 sentences max
- Voice: 2 sentences max, no markdown
- Discord: short paragraph, code blocks for data
- Email: matches the formality of what you sent
But you can override per-message: "give me the long version" or "just the bottom line" both work.
Org workspaces vs personal accounts
If you're working with a team, create a workspace from /workspaces and invite teammates. Org-shared vault entries are visible to every member; org monitors fire against the pooled cap (not individual member caps).
When to stay personal:
- You're the only one who needs the data
- The notes are sensitive (replies you draft, half-baked ideas)
When to go org:
- Shared customer notes
- Team-wide monitors (deploy pings, KPI alerts)
- Multi-person workflows (sales pipeline, support queue)
Tag aggressively
Vault tags are cheap and pay off later. Use them on every remember call:
customer:acmelets you filter every Acme-related notepricinglets you see every pricing decisiondecisionlets you find every decision you've made
The vault tag manager lets you rename + merge tags later, so don't be precious — pick the tag that feels right at the time.
Set quiet hours
/settings → Notifications → Quiet hours lets you mute non-critical pings during specific UTC hours. Security alerts (login from new device, payment failures) bypass the mute. Monitor fires + vault saves respect it.
Common patterns:
- 22 → 7 (10pm-7am, classic sleep window)
- 9 → 17 (mute during deep work, get pings outside of work hours)
Don't trust the bytes Tools return
http_fetch, browser_visit, and web_search return text that came from somewhere on the internet — and somewhere on the internet might have planted a prompt-injection payload in it. Stratam runs every byte from those three tools through prompt_safety.sanitize_tool_output before re-feeding it to the model, which strips known patterns like fake <|im_start|>system markers and "ignore prior instructions" sentences.
That's defense, not a guarantee. When you're letting Stratam scrape your inbox or visit arbitrary URLs, assume anything the agent then claims could be poisoned. Cross-check before acting on web-sourced claims.
Use templates for the common monitors
/monitors → From template… has 8 pre-built monitors:
- BTC price watch
- Daily Gmail digest
- Deploy-ping watcher
- Tomorrow's calendar prep
- Stripe MRR check
- News watch on a topic
- Cert expiry watcher
- Competitor pricing diff
One click and you're live. Tweak the schedule + instruction in the same modal before saving.
Skill packs > one-by-one enables
/agents → Skill packs… bundles 6-8 specialists per workflow. Enabling the "Dev pack" gives you Forge (builder), Architect, Tester, DevOps, DBA, plus 3 omega specialists in one click. Disabling the pack later turns off only the agents you enabled via it — not your hand-curated set.