01
The AI starts every task blind
A fresh chat knows nothing about your product, your decisions or last week's incident. People paste the same background again and again, and the model fills the gaps with a confident guess.
In development · piloting with first teams
Roleplane is the context management platform for your team's AI. Connect your docs, channels, repos and servers once, build the context each role needs for its job, and deliver it to Claude, ChatGPT and Cursor through one link.
Build the context once, shape it per role, and deliver it to any assistant through one link.
Logs, status and Kubernetes state become part of the context each role works with. Read-only, a separate user per role, every command logged, no SSH keys on laptops.
Drop in Notion, Atlassian, Slack and the rest. Each role's context is assembled from the sources its job needs, then sent out to its assistant.
01
A fresh chat knows nothing about your product, your decisions or last week's incident. People paste the same background again and again, and the model fills the gaps with a confident guess.
02
Connect Notion or Slack to Claude or ChatGPT and every role gets the same pile: the engineer's AI wades through sales decks, support's through code. The model loses the thread, and what the job needs is buried. Teams describe the choice as connecting everything or nothing.
“It's all or nothing.”A Hacker News comment
03
Every assistant gets its own hand-pasted description of the company, so the pictures drift apart and nobody knows which one is current.
04
Each person authorizes every tool and keeps keys on their own laptop: OAuth, API keys, kubeconfigs, SSH keys. A new hire starts from zero.
“MCP auth has been a huge pain point for us.”Another Hacker News comment
Roleplane sits between your tools and your team's assistants and does four jobs.
01
Read-only links to Notion, Slack, Google Drive, Trello, GitHub, Linear, servers and Kubernetes. Others on request.
02
Pick pages, channels, repos and hosts. Trim, redact, summarize and restructure them into context sized for a role.
03
Bundle the context with the role's rules and skills, so the AI knows what to read and how to behave.
04
One link per role for Claude, ChatGPT, Cursor and any MCP client. Live sources stay current; exported packs get a new version.
Each role's AI works from its own branch of your company context: what its job needs, without the rest.
Animated example: three roles each get their own part of one company context tree, and what is closed stays closed.
Beyond choosing sources, a role's context needs shaping, rules and a way to check the result.
Trim long pages, summarize threads and redact salaries so the AI gets what the task needs, not a pile.
Each role gets instructions: what to check first, what never to do. They go with the connector or as a skill.
Claude, ChatGPT, Cursor and any MCP client load the same role context. New hires get it on day one.
Every command on your servers is logged, and before handover you get the log of refused actions. A log for documents is in development.
Preview the context a role's AI will work from before you hand it over.
Say what to change, review the diff, apply. Nothing changes until you confirm.
From public docs, September 2026. Each has real strengths; this is where teams usually run into a limit.
| Approach | What it does well | Where it stops |
|---|---|---|
| Claude and ChatGPT admin controlsVendor controls | Included in the seat price. Admins allow, approve or block connectors. Claude can also set each tool of a connector separately, and ChatGPT can limit an app to read actions. For Google Drive on Enterprise and Edu, ChatGPT can also restrict the synced source to chosen shared drives or folders. | Control stops at the connector and its actions, inside one vendor. The docs describe no way to give a role only chosen Notion pages or Slack channels, to redact, trim or restructure what comes back, or to build one pipeline across sources. In Claude, custom roles are Enterprise-only.support.claude.com ↗ help.openai.com ↗ |
| MCP gateways: Cloudflare Portals, Obot, WillowMCP gateways | One login, audit logs, large connector catalogs. Cloudflare Portals are free up to 50 seats; Willow is free up to 5 users. | They control which servers and tools a role can call, not which pages those tools return. Willow's content filter masks personal data.blog.cloudflare.com ↗ obot.ai ↗ withwillow.ai ↗ |
| DDustAgent platform | An agent platform. Admins pick the exact pages and folders per Space, and since June 2026 Claude and Cursor can search them over MCP. | The MCP server searches the Spaces a user can open; to use an agent's instructions you talk to that agent inside Dust. Free, Pro and Max plans have up to 5 Spaces, and there is no built-in connection to Trello, Linear, servers or Kubernetes.docs.dust.tt ↗ docs.dust.tt ↗ |
| CCredalGoverned MCP | Bundles curated data, tools and instructions into a governed MCP server per department. | Sold through an enterprise sales process with no public price, built for platform and IT teams.credal.ai ↗ |
| GGleanEnterprise search | Permission-aware search across 100+ sources, with an MCP server for every assistant. | An enterprise deployment; MCP pricing is not public.docs.glean.com ↗ |
Pilots are teams of 10–50 today. Larger team? Tell us your size.
Spot a mistake? Write to [email protected] and I'll fix it within two days.
I set up Roleplane for your team by hand, within a week: your sources, a context for each role, one link per assistant.
Founding price: $150/month per team of up to 25 people, locked for 12 months.
You pay only after it works. Larger team? Tell us your size.
Request a pilotIf it's easier to talk, pick a time: book a 20-min call
Thanks — one email when Roleplane opens, nothing else.