The problem
A chat gives an answer. Ongoing work needs memory, roles, and recovery.
Most local AI interfaces are built for a simple exchange: a person asks a question, the model responds, and the conversation is saved. That is useful, but the person still has to track the project, choose the right model, divide the work, manage open questions, control tools, and recover after an interruption.
Cerebral adds that missing work layer. Each request belongs to a project, can be assigned to an appropriate specialist role and model, and leaves progress that can be reviewed and continued later.
How it differs
It organizes work around a project, not just a sequence of prompts.
Project + intent
The system starts with an identified body of work, its current state, constraints, and operator goal.
Role + model
A lead, helper, analyst, writer, coder, or compactor receives a clearly defined responsibility and a model that fits the available resources.
Work that continues
Jobs, questions, outputs, progress, and recovery state survive beyond a single model response.
Human acceptance
The person resolves important choices, reviews results, and decides what becomes reusable capability.
Why specialist roles matter
A smaller local model becomes more useful when each assignment has a clear purpose.
Focused roles
A smaller model can serve as a writing, analysis, coding, review, or summarization specialist within a clearly defined assignment.
Fits the computer
The system chooses models and workloads with the available memory, context, and processing capacity in mind.
Human-controlled teamwork
Specialists contribute to one project while the person retains authority over goals, tradeoffs, sensitive actions, and final acceptance.
This approach is especially useful on individual workstations and compact local systems. It supports specialization, longer assignments, and recovery without requiring every request to use the largest model or an always-connected cloud service.
Cortex integration
Cortex gives Cerebral memory, recovery, and controlled access to local tools.
Cerebral uses Cortex to recognize the active project, restore a short starting brief, find selected knowledge, preserve meaningful progress, and reach approved local actions. This keeps long-term project memory outside the model’s temporary context and prevents stored information from becoming permission to act.
Expandable by design
Add new specialists, tools, and workspaces without rebuilding the whole platform.
Cerebral keeps its core separate from agent roles, instructions, tools, reference material, and focused modules. Its existing agents can help draft a new addition, but a person still reviews and tests it before it becomes a trusted part of the system. The platform can therefore grow without silently changing itself.
Specialist roles
Add focused writing, analysis, coding, or review behavior without changing the core system.
Resources and tools
Add reusable knowledge or approved actions with clear access rules.
Modules
Add a workspace for a particular kind of work while reusing the same projects, models, tasks, and memory.
Current module
Pegasus Books turns the shared foundation into a focused long-form writing workspace.
The Books module adds planning, chapter and scene work, continuity support, specialist writing and review roles, and visible long-form progress. It reuses Cerebral’s models, agents, decisions, work tracking, and Cortex memory while adding the experience that book development needs.
Current result
An installed Windows framework used for real local-model work.
Cerebral has completed local-model requests, used structured tools, continued saved conversations, handled cancellation, coordinated specialist roles, and supported Books-module work. Those exercises demonstrate the working platform and its integration with Cortex.
Transferable capabilities
What this project demonstrates.
- AI platform architecture
- Small-model orchestration
- Semi-agent systems
- Durable jobs
- Human approval patterns
- Model resource fitting
- Tool contracts
- Module architecture
- Interruption recovery
- Cortex integration
- C# / .NET / WPF
- Local inference