Deep dive
Pick the buyer general tools ignore
General team chat is dominated by large suites, so a new product needs a clear reason to exist. Strong options include frontline teams who work on phones in shifts, regulated organisations that need on-premise or sovereign hosting, industries that need chat tied to their records, and software companies that want chat embedded in their own product for their customers.
Each of those buyers changes the MVP. Frontline teams need shift-aware notifications and simple onboarding without corporate email. Regulated buyers need retention, exports and audit logs early. Embedded chat needs an SDK and an API rather than a standalone app. Our product discovery work makes that choice explicit before the build starts.
How real-time delivery and unread state work
Each client keeps a WebSocket open to the real-time service and subscribes to the workspaces it belongs to. A new message is stored, assigned a sequence number within its channel and pushed to connected members; members who are away receive push or email according to their preferences. Clients that reconnect ask for everything after the last sequence they saw.
Unread counts and mention badges look trivial and are not. Each member has a read position per channel, updated across devices, and badges must agree on web, desktop and phone. Store read positions separately from messages and update them in batches to keep write volume manageable.
- Make message sends idempotent with a client-generated ID, so retries never duplicate messages.
- Fan out to connected clients only; let disconnected clients catch up on reconnect.
- Treat edits and deletes as events so search, previews and notifications stay consistent.
Integrations, bots and AI agents
Team chat becomes a hub when other tools can post into it and act from it. The basics are incoming webhooks, slash commands, interactive buttons and an OAuth app model with scoped permissions. Connectors for a few tools your buyers use, such as issue trackers, CRMs and calendars, deliver value on day one.
AI is the newest layer. Permission-aware search that answers questions with links to the source messages is the most useful first step, followed by recaps of busy channels. Agents that take actions in connected tools need governance: scoped credentials, approvals for sensitive actions and logs. Standards such as the Model Context Protocol make it easier to connect AI to tools consistently, and our AI agent development team builds on them.
Security, compliance and enterprise sales
Companies trust team chat with their most candid conversations, so security is part of the product. Encrypt data in transit and at rest, isolate tenants rigorously, log administrative actions and support SSO early. A SOC 2 report is a common requirement for mid-market and enterprise deals, and preparing for it shapes processes as much as code.
Compliance features follow buyers. Retention policies and exports matter for regulated industries, legal holds for litigation, data residency for European and public-sector customers and customer-managed keys for the most sensitive. Personal data in messages is covered by the GDPR and India's DPDP Act, so deletion and access requests must work across messages, files and search indexes.
Running costs
Text messages are cheap to store; files and search indexes are not. The main lines are connection servers, the database, file storage, the search cluster, push and email delivery, call infrastructure for huddles and AI inference. Free-tier limits on history and storage keep costs aligned with revenue.
Maintenance is roughly 15-20% of the build cost per year, plus infrastructure and the ongoing work of security reviews and compliance. Our application maintenance and support team usually stays involved as enterprise features are added.