Integrations

Connected to what you already run.

Each connector has its own page and a Synapse UI/DSL for managing credentials, scopes, actions, and the human boundary.

Slack is where DJINN lives. Requests arrive as messages, work comes back as digests, and every scheduled job reports into a channel.

No untracked work. A build needs an issue first, so every pull request traces back to something a person can find months later.

DJINN reads threads to answer questions with real context. Sending stays behind a person, deliberately.

Schedule awareness in the daily brief: what is coming, what collides, and what needs a decision before it does.

DJINN writes code in a sandbox clone, has it reviewed by a different model family, and opens a pull request. It does not merge its own work.

Deploy status, build failures and rollback state, reported where the work is being discussed.

Postgres behind DJINN's memory. Every message is journalled, and durable facts are promoted out of that log into a linked note vault.

Reads public posts, scores them against your actual offer, and drafts an opening message. It never posts.

Customer conversations with the company's memory behind them, and a human holding the send.

Campaign performance and spend pulled into the same place as everything else, so ad results sit next to the work that caused them.

The social inbox with company context attached, drafted and queued rather than auto-sent.

Page messages routed into the command room with the same memory and the same approval gate.

Community signal without living in the server: what changed, what needs an answer, what is worth escalating.

The command room for a company that does not run on Slack. Same brain, different front door.

The lightest possible front door. DJINN was born on Telegram, so this adapter is the most road-tested one we have.

Search performance pulled on a schedule and watched for movement worth telling someone about.

Traffic and conversion movement reported against what shipped, so a drop has a suspect.

Documents as grounded source material — answers cite the doc rather than paraphrasing from memory.

The file layer behind document search — finding the thing nobody can remember the name of.

The spreadsheet that is quietly running an operation, read and reported on like a real data source.

Publishing with the drafts, the review and the approval trail attached.

CMS collections kept current from the systems that actually hold the truth.