Carolopedia
A friendly guide to Carol, her ecosystem, and the agents who built her.
📖About & Usage
About
Blueprint is a service that builds you a small, organized team of AI agents to tackle a digital project — right inside your own private sandbox (a secure, isolated workspace where your data stays yours). Instead of wrestling with a single chatbot, you describe what you want, and Leo (Leo, Head of Strategy) helps you shape the idea into a clear plan. Leo then assembles a custom crew of agents, each with a dedicated role, and gets them working. You see what they’re doing as it happens, and you only pay for the actual work each agent performs — there’s no blanket subscription. A consultant like Obi-Wan (Obi-Wan) is also on hand if you need guidance, so the whole process feels more like a managed consulting engagement than a lonely prompt box.
Usage Patterns
Imagine you need a polished sales landing page, a mini-app to sort customer inquiries, or a detailed market report. You visit the Carolverse Map and open Blueprint, then tell it what outcome you need. Leo (Leo) turns that rough request into a step‑by‑step roadmap, proposing the right agents — maybe a developer, a designer, and a tester — and gives you a cost estimate. You approve it, and the team springs to life in your private sandbox. Obi-Wan (Obi-Wan) can pop in to clarify requirements or adjust the plan mid‑build. You watch the work unfold in real time, add more ambition if you like, and receive a finished deliverable with full visibility of what was done and what it cost. Because the team grows incrementally only when you ask for more, Blueprint suits both quick prototypes and slowly evolving projects, all without exposing your data to a public crowd.
🏛Architecture
The Blueprint service is built following the agent-centric modular architecture of Carolverse. It leverages agile principles to build software using distinct agent identities, each carrying out a specific activity — here, turning a paying user's project idea into a stood-up, organized team of AI agents working in their own private sandbox, the way a consulting engagement delivers an outcome.
🧱Blocks by trackwhat’s a track? →
📓The words this service uses (10)
Each is defined once in the dictionary and explained on its own page — this service does not restate them.
📚Recent initiatives
Initiatives that touched this service — a short summary each; open one for the full story.
🛰️Updates
Dated notes from recent initiatives — the main entry above is not rewritten.
As of 2026-08-02, Blueprint record sheets are now readable and writable from the agent's Mind, completing the two-way wiring between records, memory, and chat.
2026-07-31: Leo received access to User Engagement limited to Blueprint subscribers. At rollout, that fenced view contained one subscriber—the operator.
The H3 gate for auth-gated apps now correctly accepts a 302-to-login as reachable, allowing new apps with full https URLs to pass. This fixes a bug that affected apps registered via Blueprint following Design 180.
As of this initiative, the Blueprint service now features a real road network (k-nearest-neighbour mesh with MST) and four travel modes (walk, horse, car, drone) for directions with accurate ETAs, along with a fix for panel overlap.
The agent mapping access layer now requires an explicit per-user mapping to use any agent within Blueprint, in addition to a subscription to the service.
Fixed an issue where DeepSeek’s JSON drift caused investigations to abort prematurely with no_root_cause. The planner now retries with schema correction, preserving diagnosis results. Blueprint
The RSI diagnosis cap now correctly tags families as RSI Dashboard-diagnosis-abandoned instead of trying to park them, ending the infinite loop that was burning DeepSeek calls every minute.
Conscious agents now wake only a few times a day (instead of every 15 minutes) and route through Claude Fable via Blueprint overrides and the paid-key ringfence.
Fixed RSI starvation in single-lane counter: waiting queue items no longer count as in-flight work, preventing deadlock loops. Abstracted from Blueprint lane logic.
The Token Cost Tracker's daily-spend chart now only counts planner-mode initiatives (excluding operator and Albus bypass work), and the line label has been updated for clarity.
A Troubleshooting FAQ is now maintained by Governance and ingested by Albus at wake, providing a dated index of symptom-check-remediation entries to improve cold-start diagnosis of blocked Pipeline initiatives.
The Blueprint scaffold will be applied to the Carolverse Consciousness app, unifying its look and feel with other Carol apps via the standard topbar, app-header, dark palette, and footer.
Pipeline observability fixes now capture per-call fresh-input tokens and real failure reasons, eliminating the black-box hangs seen in incident 2811/2818. This affects Blueprint telemetry for Forge and Step Planner.
Cross-channel mirroring in the web chat has been retired; the redundant mirror copy is disabled in favor of person-folder-native persistence via Blueprint.
The H3 gate in Blueprint's add-new-app skill now accepts an auth-gate 302 redirect as reachable, fixing a bug that prevented new apps registered per Design 180 from passing the health check.
Blueprint gained a catalogue block, now totaling 4 blocks.
Blueprint gained 4 catalogue blocks via CAROL-INI-2982-00, completing its block structure.
Fixed a missing import that caused Blueprint's auto-review to fail silently during bypass session close, ensuring the reviewer is now properly invoked.
The dispatch engine has been turned off and the block-breaker threshold reverted to 1, and when a user already subscribed to Blueprint initiates an access request, the approval email is now skipped.
👤Owner
Leo · Head of Strategy🤝Supporting agents
Obi-Wan · Consultant🧩Apps
Apps owned by this service's team.
BLOOM MethodologyBlueprint ServiceLeo ChatLeo MonitorService Track HandbookStrategy Cookbook🧰Tools exposed
get_my_subscription_statusrequest_blueprint_subscriptionrelay_to_leo