Keep team materials and decisions in one project. Review the background sent to each AI, submitted results and the changes your team chooses.
Start with a two-week pilot on one team — everyone keeps the AI they already use.
Individual harnesses already share rules, inspect context, enforce permissions and run agent loops. The gap is not missing capability inside one tool; it is the lack of a shared standard for which approved project version crossed into multiple harnesses, and why.
One harness can show what its own session loaded. A separate ledger is still needed to show which approved version actually reached whom across Claude, Cursor and Codex.
Repository rules are strong for code, while project background also lives in meetings, plans and design. Updating each tool file separately makes the team reconcile which version is the approved current background.
Teams can solve this with structured docs, linters, hooks, permissions and recurring cleanup. Across multiple harnesses and non-repository sources, someone must still design and maintain that system.
Run AI in the tools you use. Manage approved background and delivery history in DroFi. Spend less effort remembering which version went to each tool.
CLAUDE.md, Project Rules and Team Rules, context views, hooks and permissions, and agent loops belong to existing harnesses. DroFi has no reason to recreate them.
As OpenAI's Harness Engineering shows, a team can make repository docs the source of truth and add mechanical checks plus recurring cleanup. The design and maintenance burden stays with the internal platform team.
DroFi selects blocks, documents and files from approved meeting, planning, design and code context; shows inclusion, omission and fingerprints before sending; then records only delivery actually routed through DroFi.
If you have two or more harnesses + project knowledge outside the repository + export control or evidence requirements, DroFi is the packaged alternative to building this layer yourself. For a solo, single-harness, repository-only workflow, the harness may be enough.
Official baseline: OpenAI Harness Engineering ↗ · Claude Code memory & rules ↗ · Claude Code context ↗ · Cursor Rules ↗
Review changes, content to send and each AI’s delivery history in one place. See the implemented UI →
Documents, decisions and guidelines live together per project. An AI connected over MCP asks again when work starts and gets the latest source of truth at that moment; copies and files are labeled snapshots.
MCP project context is scoped to the recipient role; a task prompt is scoped to the selected blocks and task itself. DroFi keeps those two scopes explicit.
Blocks and documents marked no-export are left out of AI context DroFi creates. They are not filtered after the fact — what must not leave is never packed in the first place.
DroFi records the actor, channel, exact included blocks and size for context it delivers. It does not claim visibility into material pasted directly outside DroFi.
Inheriting org guidelines into every project and preventing local overrides is being designed through pilots. Project-level shared guidelines are available today.
Today each member can see their own monthly AI usage and plan limit. Organization-wide aggregation and administrator-set caps are being shaped as enterprise pilot requirements.
This is not a tooling migration. Everyone keeps the AI they already use; what changes is where the context is governed.
Bring in the documents and repositories of a project already in flight and you're running. Three things get tested here: does output quality even out, is what left visible in the record, and does no-export actually hold.
Promotion into org-wide guidelines and role-level context policy are currently being designed through pilots. First validate requirements with today's project guidelines, roles and no-export controls, then define the organization policy with security — prove what should leave before scaling it.
Reusing org guidelines, consolidated usage and delivery auditing, and company-wide onboarding are planned areas to validate after the pilot. Today, prove the rollout with one team's project source of truth and project delivery history.
Start normal collaboration on Team. If you need org auditing, security review or self-hosting, contact us about an Enterprise pilot that shapes those planned features. Compare all plans →
Two weeks with one team is enough to judge it. Tell us your scope and security requirements and we'll take it from there.