How It Works
Logic Context generates a BRD with a fleet of specialist agents coordinated by an orchestrator. Each section of the document is owned by a specialist that knows how to read one artifact type and how to write it up. The specialists run in parallel, an optional reviewer grades each section, and a synthesis step assembles the finished document. Results stream to the browser as each section completes.
The generation pipeline
- Start. You pick a project and the sections you want. The API returns a
jobIdimmediately and runs the work in the background, so there is no long-held request. - Fan out. The orchestrator resolves the project and starts the specialists in parallel. Each reads its artifacts through the Flowon MCP server, on behalf of the signed-in user, and writes its section.
- Review. If the quality gate is on, a reviewer scores each finished section against section-specific criteria and sends weak sections back to be regenerated with its critique.
- Stream. Each section is saved and emitted the moment it is ready, so the document renders progressively instead of appearing all at once. Refreshing the page reconnects to the same job and replays completed sections.
- Synthesize. Once every section is done, a synthesis step assembles the ordered, ID-stamped document and computes cross-section references.
- Store. The finished BRD is saved to history, ready for chat refinement and export.
Architecture
The orchestrator fans out eleven specialists in parallel. Each finished section passes through the reviewer, which either accepts it or sends it back for regeneration. Synthesis assembles the accepted sections. The router handles chat refinement by targeting specific specialists.
The agent fleet
| Agent | Role |
|---|---|
| OrchestratorAgent | Resolves the project, fans out the specialists, and drives the quality-gate loop |
| 11 specialist agents | One per section, each reads its artifact type and writes that section |
| ReviewerAgent | Scores each finished section against section-specific criteria and drives regeneration |
| RouterAgent | During chat refinement, decides which section a natural-language request targets |
| SynthesisAgent | Assembles the sections into one ordered, ID-stamped, cross-referenced document |
The eleven specialists and the sections they write are detailed in The BRD Sections.
Why parallel specialists
A single model asked to document an entire project at once produces shallow, uneven output and is slow. Logic Context gives each artifact type its own specialist with a focused job and focused knowledge of that type. The specialists run at the same time, so a full document is produced far faster than a sequential pass, and each section is written by an agent that only has to be good at one thing.
The two-phase business-process treatment
Business processes are the hardest artifact to document well, because a single process definition can be large and carries both structure and behavior. Logic Context handles them in two phases:
- Extraction. The raw process definition is distilled into its stages, transitions, actors, and SLAs.
- Insight. A second pass reads that compact structure and derives process-mining insights: the personas involved, the happy path and the critical path, the decision points, automation coverage, and bottleneck risks.
Both phases feed the Business Processes section, so the document describes not just what the process is but how it behaves and where its risks lie. This is covered further in The BRD Sections.
Grounded and traceable
Every specialist writes only from what it reads through the MCP server, and every generated requirement carries a traceability link back to the Flowon artifact it came from. The document is business-readable on the surface and fully auditable underneath.