How it Works

Multi-Agent AI Pipeline

Traceformer.io uses a three-phase orchestrator pipeline: planning, parallel evidence retrieval, and synthesis. This structure maximizes review quality while minimizing token and model cost.

The Review Pipeline

Schematic

Datasheets

Up to 40

1. Planner

Analyzes schematic, creates work order

2. Workers

Up to 10 parallel agents

W1
W2
W3
W4
W5
W6
W7
W8
W9
W10

3. Merger

Synthesizes evidence into review

2 Errors
6 Warnings
18 Verified
7 Missing Info

Planner

  • Parses your complete KiCad project or Altium netlist file
  • Identifies circuit subsystems (power, comms, sensors, etc.)
  • Creates evidence requirements for each subsystem

Workers

  • Run in parallel for speed (up to 10 simultaneous)
  • Each specializes in one subsystem
  • Search datasheets for relevant specs
  • Use lightweight models to minimize cost

Merger

  • Receives all evidence from workers
  • Synthesizes findings into structured output
  • Produces actionable recommendations

Why This Architecture?

Cost Efficiency

Pick between frontier reasoning models such as GPT 5.2, Opus 4.5, or Sonnet 4.5 for planning and synthesis. The bulk of the work (evidence gathering) uses lightweight, fast models that cost a fraction of the price.

Speed Through Parallelism

Workers execute simultaneously, reducing wall-clock time significantly compared to sequential processing. Your review completes faster.

Fault-Tolerant Analysis

If a worker fails to gather useful evidence, the model does not get hung up. You still receive a useful review even when a worker cannot find relevant data, and the system clearly reports what it coulden't verify.

How it Works | Traceformer.io