zkagi

Start by deciding what zkagi is allowed to know and do. In the setup screen, pick where it runs (your machine or your own server), then connect only the sources you want: a project folder, a notes directory, a calendar file, or a mailbox. Define scopes per workspace—what files it can read, for how long, and whether it can keep short-term memory. Turn on redaction rules for names, emails, and IDs. Every action is consented and logged, and you can wipe sessions, export transcripts, or rotate keys in a click. This puts you in charge of context, retention, and access before any prompt is sent.

For writing, create a workspace per deliverable—blog post, whitepaper, landing page, or report. Drop in your brief, brand voice notes, and any reference PDFs. Ask zkagi to outline three structures, compare them to your style guide, then expand the winner into a first draft with source-backed facts. Use “quote-only” mode to force citations from your attached docs. Iterate with targeted edits: tighten the lead, add an example from page 8, swap jargon for plain language. Generate variations for different audiences, and export to Markdown or DOCX with footnotes preserved. Because the model works within your defined scope, it won’t leak external data or pull in unwanted sources.

For planning and research, spin up a board and paste articles, meeting notes, and datasets. Tag priorities and deadlines. Ask zkagi to synthesize findings into a brief with callouts for risks, assumptions, and open questions. Convert insights into an action plan—tasks, owners, and checkpoints—then sync the timeline to an ICS calendar. Use “evidence view” to drill into where each claim came from. Share an encrypted snapshot with a client or teammate; they can review locally without exposing your full workspace. When an initiative is complete, archive the board with a signed summary and a decision log for audit and future handoffs.

For coding and data work, connect a repo and point zkagi at specific directories. It can read code in-context to answer questions, propose refactors, or create tests—without sending your source elsewhere. Use the secrets vault for tokens; enable ephemeral context so keys never persist in prompts. Run static analysis suggestions, generate migration steps, or scaffold an internal CLI. For analysis, attach CSV or Parquet files and ask targeted queries; export notebooks with the exact prompts and operations attached for reproducibility. When you need automation, save your steps as a workflow you can run from the CLI or on a schedule, always constrained by the permissions you set.