Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.sigilix.ai/llms.txt

Use this file to discover all available pages before exploring further.

Most AI code review tools run a single model with a single prompt over the diff. The model is asked to be everything at once — security expert, performance engineer, architect, semantics nitpicker — and it does each role about half as well as a focused specialist would. Sigilix’s architecture is different. Five agents run in sequence: four parallel specialists, each tuned for one class of failure, and a synthesizer that arbitrates between them.

​
The topology

Every review goes:
  1. Specialists run in parallel — Glyph, Warden, Spark, and Weave receive the same diff but with different prompts and different model choices. They can’t see each other’s findings.
  2. Findings flow into Core — the synthesizer sees all four streams plus the diff itself.
  3. Core deduplicates, calibrates, and renders — overlapping findings collapse into one; severity shifts based on the agreement signal; the final verdict is decided.
  4. One comment is posted — single GitHub review with the synthesizer summary at the top and inline findings below.

​
Why this beats single-agent review

​
1. Different prompts catch different things

A single-agent reviewer with one prompt can ask the model to “look for security issues, performance issues, architectural violations, and naming problems.” The model attends to roughly one of those at a time and trades off depth. Sigilix’s specialists each have a focused prompt. Warden is asked only about security. Its prompt is 600 words of OWASP-relevant patterns, secret-leak heuristics, and authentication boundary rules. The model running Warden’s prompt finds more security issues than the same model running a generalist prompt — by a wide margin.

​
2. Different models suit different roles

Sigilix uses different model providers for different specialists. Logic-heavy tasks like Glyph’s architectural reasoning use deepseek-v4-pro for its proof-style chain-of-thought. Security tasks where false-positive cost is high use kimi-k2.5 with retry+fallback to a different provider. Synthesis uses kimi-k2.6 for its calibration ability. The right model for the job, not one model for everything. See Specialists for per-role model selection.

​
3. Cross-reference suppresses hallucinations

Single-agent review hallucinates findings. The model is confidently wrong about a function being unused, a variable being uninitialized, or a security pattern being broken — when the reviewer reads the file in question, the finding is fiction. Sigilix’s synthesizer cross-references findings with the source code. If Warden flags a SQL injection at line 42 but Core’s structural-provenance check shows the parameter actually passes through a parameterized-query helper, the finding is suppressed before it reaches you. The cross-reference is the difference between “AI review you tolerate” and “AI review you trust.”

​
4. Severity calibration uses the agreement signal

When multiple specialists flag the same code, that’s a strong signal. Core escalates the severity in those cases:
  • One specialist flags + low confidence → Info
  • One specialist flags + high confidence → Warning
  • Two+ specialists flag → Warning or Critical (depending on category)
  • Specialist + Core’s structural check confirms → Critical
A single-agent reviewer can’t calibrate this way — it has nobody to disagree with.

​
5. The interface is one comment, not 40

If you’ve used a single-agent reviewer that dumps every thought it has into the PR thread, you know the cost. Reviewers stop reading after the third “Consider adding a docstring.” Real findings get buried. Core deduplicates relentlessly. If Warden and Spark both flag the same loop, you see one comment, not two. If a finding is a duplicate of one already posted on a prior SHA, you see it once.

​
The trade-off

Multi-agent review is more expensive than single-agent review. Five model calls per PR cost more than one. Sigilix’s pricing reflects that — see the pricing tiers on the marketing site. For most teams, the trade-off is worth it: a single missed security bug shipped to production costs vastly more than the per-PR review cost. For teams with very high PR volume, the per-PR cost can be tuned via rate limits and specialist disabling.

Specialists

Each of the four specialists in detail — what they catch, sample findings, model selection.

Synthesizer

Core’s pipeline: collect → cross-reference → calibrate → render.