Dreamer

Your agents learn as a team.

Dreamer collects what Claude Code, Cursor, Codex and other coding agents learn at work and turns it into context every new session starts with.

Works with any agent that speaks MCP, including

  • Claude Code
  • Cursor
  • Codex
  • GitHub Copilot
dreaming
AGENTS.md

## Learned by the team

  • · Prices are stored in cents.

+Active customer: paid in 90 days.

+Tests use the billing sandbox.

+Migrations need a lock timeout.

notenotenote

The same lesson, every week.

A correction lives in one chat with one agent. When the session ends, it is gone, and the next engineer pays for it again.

  1. MonAn active customer is one who paid in the last 90 days.
  2. TueActive means paid in the last 90 days, not logged in.×2
  3. WedAgain: active customers paid within 90 days.×3
  4. ThuLogins don’t count. Active = paid in 90 days.×4
  5. FriActive customer = paid within 90 days. Please remember.×5
MonWith Dreamer, a new engineer's agentUsing the team's definition: paid in the last 90 days.

One correction, every agent.

Agents connect to Dreamer over MCP, so engineers keep the tools they already use. A lesson learned in Cursor reaches Claude Code the next morning.

  1. During the day

    Claude Code

    Active means paid in 90 days, not logged in.

    Noted for the team.

    An engineer corrects an agent. The agent saves it as a note for the team.

  2. Overnight

    your note2 similar notesold rule

    Active customer: paid in the last 90 days.

    Dreamer merges the team’s notes into one rule and retires the rules it replaces.

  3. Next morning

    Team context updateapproved
    AGENTS.md+ Active customer: paid in the last 90 days.− Active customer: logged in within 30 days.

    From 3 notes: Claude Code, Cursor and Codex

    Your team approves the change. Every new session starts with it.

Every lesson keeps paying off.

Spend tokens on the task, not on rediscovery

Without shared context, agents read through the codebase to rediscover rules the team already knows. Dreamer gives them those rules up front and keeps the file lean: only what the code can’t tell them, with stale rules removed.

Tokens one agent session spends

Without Dreamer

rediscoverythe actual task

With Dreamer

the actual tasksaved

See what your agents learned

A weekly digest shows new rules, retired ones and what agents rely on most.

This week
  • New rules+14
  • Retired−3
  • Most usedActive customer

Knowledge stays when people leave

What a departing engineer taught their agents keeps working for the team.

Left in March

12 rules in use

You decide what agents learn.

Dreamer runs in our EU cloud or in your own, within a spend cap you set. Every change to what your agents read is reviewed, logged and reversible.

Workspace settings
  • Review before rollout
  • Filter secrets and personal data
  • Shared memoryOrganisationData teamBilling team
  • Monthly spend cap€150
  • HostingEU cloudYour cloud
  • Audit logevery change, with who approved it

Questions before a pilot

Which coding agents does Dreamer work with?

Claude Code, Cursor, Codex, GitHub Copilot and any other agent that supports MCP. Teams that mix tools share one memory, so a lesson learned in Cursor reaches Claude Code too.

Do our engineers have to change how they work?

No. Connecting an agent is one entry in its configuration. Agents decide when something is worth noting, engineers keep working as before, and someone on the team reviews the update when it arrives.

How is this different from the memory built into Claude Code or Cursor?

Built-in memory stays inside one tool and usually one person’s machine. Dreamer pools what every agent on the team learns, merges and reviews it, and ships it as the AGENTS.md and skills files that all of your agents already read.

Will it increase our token costs?

Consolidation runs on a schedule, within a monthly cap you set. In return, agents start with rules they would otherwise rediscover by reading code, and the context stays short because duplicates are merged and outdated rules are removed.

Where is our data stored?

In our EU cloud, or in your own cloud account if you prefer. Secrets and personal data are filtered out before anything is stored, and every change is logged.

What stops a wrong lesson from spreading?

Updates are reviewed before they reach your agents. Agents also report when a rule turns out to be wrong, and Dreamer retires it with a record of why. Rules you pin are never changed automatically.

We already maintain AGENTS.md files. What happens to them?

Dreamer imports them as the starting point and keeps what you pin. It builds on your files instead of replacing them.

How long does it take to get started?

A first team can be running within a day. On the intro call we look at how your team uses agents today and plan a pilot around one or two repositories.

What does it cost?

Pricing depends on team size. We are onboarding a limited number of teams right now; book an intro call to talk about a pilot.

Let your agents sleep on it.

Book an intro call