Retrieval-grade knowledge, with a test loop

The knowledge layer
your AI agents
can actually
trust.

Gyria sits between your source systems and your AI. It ingests everything, answers with grounded citations, and continuously proves the answers are right, so retrieval stops being a guess.

Sign inSee the API
500 free credits · no card · REST + MCP included
gyria · answerlive
Q: What's our refund window for enterprise?
Refund policy v7Enterprise addendumBilling FAQ
hybrid retrieval · rerank · auto-filter (tier=enterprise)
Enterprise customers have a 60-day refund window1, processed to the original payment method within 5 business days2.
Faithful 0.98218 msgrounded · 3 chunks
ZendeskConfluenceServiceNowSalesforceSharePointWeb crawlPDF & DOCXQ&A pairsImagesYour APIZendeskConfluenceServiceNowSalesforceSharePointWeb crawlPDF & DOCXQ&A pairsImagesYour API
<80ms
vector search at 30k chunks
0.98
faithfulness, scored on every answer
100%
of answers grounded in a real, cited source
500
free credits to start
How it works

Ingest. Retrieve. Watch.

STEP 1

Ingest

Connect sources or push docs over the API. Converted to markdown, chunked, contextualized, embedded, incrementally and on a schedule.

STEP 2

Retrieve

Grounded, cited answers with the exact chunks. Hybrid retrieval, rerank, auto-filters, agentic decomposition, and conversational memory.

STEP 3

Watch

Freshness scored, contradictions and gaps flagged, and answer quality proven by a continuous, judged test suite.

Everything included

Built for production RAG

Hybrid retrieval

Lexical + dense vector search fused with Reciprocal Rank Fusion over a 1024-dim index. Exact terms and paraphrases both land, in tens of milliseconds at scale.

Agentic retrieval

Decomposes complex questions into sub-questions, retrieves each, and merges.

Auto-filter

Infers metadata filters from the query, validated against your real vocabulary.

A real test loop

Golden questions, AI judging, and a faithfulness / precision / recall panel with a hallucination flag. Catch regressions the moment knowledge changes.

Permission-aware

Per-source ACLs enforced at retrieval. Answers never leak a doc the caller cannot see.

Contextual + rerank

Chunks are situated in their document before embedding, with an optional LLM rerank.

Headless control plane

A full REST + MCP management API: ingest, author, configure, delete. Run Gyria with no dashboard at all and let your AI agent operate the entire workspace.

Every source

Crawl sites, upload PDFs and DOCX (images captioned too), or connect Zendesk, Confluence, ServiceNow, Salesforce and SharePoint, on a schedule.

Agentic retrieval

Hard questions, unravelled and rewoven.

A compound question rarely lives in one chunk. Gyria decomposes it into sub-questions, retrieves each on its own, then merges the threads into a single grounded answer with every citation intact.

Automatic multi-hop decompositionSub-answers fused, citations preservedConversational memory across turns
How does the enterprise refund window differ from standard, and who approves it?
Enterprise window
v7 · 60d
Standard window
FAQ · 30d
Approval owner
RACI · Finance
Merged grounded answerFaithful 0.98
Not just retrieval

A radar that tells you when the knowledge is wrong.

Most RAG stacks answer and hope. Gyria runs a continuous, judged test loop and scores every answer, so a bad document surfaces as a finding, not as a wrong answer to your customer.

Faithfulness, relevancy, precision & recall scored per answer
Contradictions, gaps, and stale docs surfaced automatically
Thumbs-down feedback becomes a tracked test case
For developers

One key. Fully headless.

A single workspace API key drives everything over REST and MCP: ingest and author content, manage sources, metadata, schedules and permissions, and ask grounded questions. Gyria runs with no dashboard at all, so your own AI agent can operate the whole knowledge base for you.

Run it fully headless — no dashboard requiredAn agent drives the whole loop: ingest, curate, evaluateMCP-native — point Claude or any agent at one endpoint
Get your API key
curl -X POST $HOST/api/v1/answer \
  -H "Authorization: Bearer gyr_live_..." \
  -H "content-type: application/json" \
  -d '{"question":"refund window for enterprise?",
       "agentic":true, "autoFilter":true}'

# grounded answer + inline citations,
# a trace of every retrieval step, and
# an eval-ready faithfulness score.

Put a radar on your knowledge.

Spin up a workspace, connect a source, and ask your first grounded question in minutes. 500 credits on the house.

Sign in