Multi-agent requirements engineering

Requirements youcan specify

Turn raw stakeholder text into a complete requirements specification — in IEEE 830, ISO 29148, Volere, or Agile PRD format — drafted, reviewed, and refined by a multi-agent AI pipeline, then verified against your real codebase.

5agents in the pipelineMULTI-AGENT
4LLM providers, your keyBRING YOUR OWN KEY
4spec formats: IEEE · ISO · Volere · PRDMULTI-FORMAT
2max reflection passes per draftQUALITY LOOP
5agents in the pipelineMULTI-AGENT
4LLM providers, your keyBRING YOUR OWN KEY
4spec formats: IEEE · ISO · Volere · PRDMULTI-FORMAT
2max reflection passes per draftQUALITY LOOP
Capabilities

From stakeholder text
to verified specification.

01

Multi-Format SRS Generation

Raw stakeholder text becomes a full specification in the standard you pick — IEEE 830, ISO/IEC/IEEE 29148, Volere, or Agile PRD. The chosen format drives the real structure end-to-end, drafted by a Product Owner, Architect, and Developer agent chain.

02

Quality Reflection Loop

A Reviewer and a Critic agent grade every draft against a 6-dimension quality rubric and send it back for surgical refinement — up to two passes — before it ever reaches you.

03

Bring Your Own LLM Key

Choose Gemini, OpenAI, Claude, or Grok and generate on your own API key — encrypted at rest and used only for your analyses. Keys stay yours; the platform key is reserved for embeddings.

AB
04

Live Diagrams

Data-flow and sequence diagrams render live from Mermaid, with AI self-repair for broken syntax before they're ever shown to you.

05

Code Traceability CLI

The sra CLI pulls your finalized spec, checks which functional requirements have real files backing them in your source tree, and pushes verification results back to the platform.

$ sra sync$ sra check$ sra push
Process

Three steps.
One verified spec.

request.json
1POST /api/analysis
2
3{
4  "projectName": "Inventory Manager",
5  "rawText": "Warehouse staff need to scan
6    incoming stock, flag low quantities,
7    and generate reorder reports..."
8}
Ready
Architecture

Grounded by
design.

Every draft is grounded in retrieved context, not just model guesswork — and gated by a reflection loop before it reaches you. Nothing ships without passing the quality bar.

5
Agents in the pipeline
2
Max reflection passes
85
Quality threshold / 100
Analysis PipelineSequential
Product Owner
Refines scope & features
Stage 1
RAG Retrieval
pgvector cosine similarity
Stage 2
Architect
Designs with retrieved context
Stage 3
Developer
Drafts SRS sections
Stage 4
Reviewer
Approve / reject gate
Stage 5
Critic
6-dimension quality score
Stage 6
Under the hood

A pipeline you
can inspect.

0
AI agents in the pipeline
0
LLM providers supported
0
Max reflection passes per draft
0 / 100
Quality score threshold
Integrations

Real infrastructure.
No smoke and mirrors.

The providers and services SRA actually talks to, end to end.

Gemini
LLM Provider
OpenAI
LLM Provider
Claude
LLM Provider
Grok
LLM Provider
Google
OAuth
GitHub
OAuth
PostgreSQL
Database + pgvector
Redis
Rate Limiting + Pub/Sub
QStash
Job Queue
Gemini
LLM Provider
OpenAI
LLM Provider
Claude
LLM Provider
Grok
LLM Provider
Google
OAuth
GitHub
OAuth
PostgreSQL
Database + pgvector
Redis
Rate Limiting + Pub/Sub
QStash
Job Queue
QStash
Job Queue
Redis
Rate Limiting + Pub/Sub
PostgreSQL
Database + pgvector
GitHub
OAuth
Google
OAuth
Grok
LLM Provider
Claude
LLM Provider
OpenAI
LLM Provider
Gemini
LLM Provider
QStash
Job Queue
Redis
Rate Limiting + Pub/Sub
PostgreSQL
Database + pgvector
GitHub
OAuth
Google
OAuth
Grok
LLM Provider
Claude
LLM Provider
OpenAI
LLM Provider
Gemini
LLM Provider
Security

Your keys.
Your data.

You bring the API key, we handle it carefully. Encryption, session security, and rate limiting are built into every layer of the platform — not bolted on after the fact.

JWTOAuth 2.0AES-256-GCMCSRF ProtectedRate Limited

Encrypted provider keys

Your OpenAI, Claude, Grok, or Gemini API key is encrypted at rest with AES-256-GCM and never returned in plaintext.

OAuth + JWT sessions

Sign in with Google or GitHub through a CSRF-protected, one-time exchange-code flow — no tokens ever land in a redirect URL.

Distributed rate limiting

A Redis-backed sliding-window limiter protects auth and AI endpoints across every backend replica.

Full audit trail

Sensitive actions are recorded through dedicated audit-logging middleware for traceability.

For developers

Specs shouldn't
go stale.

The sra CLI closes the loop between the spec and your repo — sync the finalized requirements, check what's actually implemented, and push the results back.

Zero-config init

sra init writes a gitignored config file for you — no manual setup.

Existence-checked

check confirms each requirement's referenced files exist — an honest, fast signal, not a claim of correctness.

Round-trips results

push sends verification status back so it's visible on the analysis timeline.

doctor built in

One command to diagnose env and connectivity issues before you file a bug.

npm install -g @sra-srs/sra-cli
# verify it's on your PATH
sra doctor
@sra-srs/sra-cli

Ready to turn notes
into a real spec?

Paste your stakeholder text, pick a provider and format, and watch the pipeline draft, review, and refine a standards-compliant spec you can verify against real code.