BrainSoup

BrainSoup enables users to create and manage multiple AI agents with distinct roles and responsibilities that work collaboratively on complex tasks. The system uses a role-definition framework where each agent is configured with specific instructions, capabilities, and behavioral constraints, then coordinates their execution through a task queue and inter-agent messaging system. Agents can hand off work to each other based on task requirements, enabling hierarchical problem decomposition without requiring manual workflow definition.

BrainSoup provides a unified interface for connecting to multiple LLM providers (OpenAI, Anthropic, local models) through an abstraction layer that normalizes API differences and handles provider-specific authentication. The system maintains connection pooling and request queuing to manage concurrent agent requests across different backends, allowing users to route different agents to different models based on cost, latency, or capability requirements.

BrainSoup implements automatic error detection and recovery mechanisms for failed agent tasks, including configurable retry strategies with exponential backoff, fallback agent assignment, and manual intervention workflows. The system captures error context and provides detailed failure reports to help users understand why tasks failed and how to resolve issues.

BrainSoup tracks token usage and API costs across all agent executions, providing per-agent and per-task cost breakdowns. The system enables users to set cost budgets, monitor spending in real-time, and identify optimization opportunities (e.g., using cheaper models for simple tasks). Cost data is aggregated and visualized to help users understand their LLM spending patterns.

BrainSoup maintains agent-specific memory stores that persist across sessions, enabling agents to retain knowledge from previous interactions and build context over time. The system implements a hybrid memory architecture combining short-term conversation context (in-memory for current session) with long-term knowledge storage (persisted to disk), allowing agents to reference past decisions and accumulated information without manual context injection.

BrainSoup analyzes complex user requests and automatically breaks them into subtasks that can be distributed across the agent team, with dependency tracking and execution ordering. The system uses a planning engine that builds a directed acyclic graph (DAG) of task dependencies, identifies parallelizable work, and sequences execution to minimize total completion time while respecting data dependencies between subtasks.

BrainSoup allows users to define and modify agent behavior through a system prompt and instruction framework, where each agent can be configured with specific guidelines, constraints, and behavioral patterns. The system supports instruction versioning and templates, enabling users to create agent archetypes (researcher, writer, analyst) that can be instantiated with domain-specific customizations without code changes.

BrainSoup provides real-time visibility into agent execution through comprehensive logging of all agent actions, decisions, and outputs. The system captures execution traces including LLM prompts, responses, token usage, and timing information, storing them in a queryable log that enables debugging, auditing, and performance analysis of agent workflows.

BrainSoup enables agents to read, analyze, and process files and documents from the local filesystem, with support for multiple formats (text, PDF, images, code). The system provides agents with file I/O capabilities through a sandboxed interface that prevents unauthorized access while allowing agents to work with project documents, code repositories, and data files as part of their tasks.

BrainSoup provides a graphical interface for creating, configuring, and managing multiple agents without requiring code. The UI enables users to define agent roles, assign capabilities, set behavioral parameters, and organize agents into teams through visual configuration tools. The system stores agent configurations persistently, allowing users to save, load, and modify agent teams across sessions.

BrainSoup implements a message-passing system that allows agents to communicate with each other, request assistance, and share results. Agents can query other agents' outputs, ask for specialized help, and coordinate on complex tasks through a structured communication protocol that maintains context and prevents circular dependencies.

BrainSoup supports scheduling agents to run tasks on a recurring basis (hourly, daily, weekly) or triggering execution based on external events or file system changes. The system maintains a task scheduler that manages agent execution timing, handles missed runs, and provides retry logic for failed tasks. Users can define trigger conditions and execution schedules through the UI without requiring cron syntax or code.