the best and fastest
agentic search models
for private data

SID trains search swarms that find better documents, emails, research, and memories than Opus and Astra, while being 8x faster and 20x more cost efficient.

>1T tokens in production

Frontier quality, at a fraction of the time and cost.

See the evals
  • The Bitter Lesson for search

    Traditional search pipelines (“RAG”) drop off rapidly as datasets get bigger or the questions harder. Their only way to improve is adding more steps, paths, and complexity.

    SID Pipelines Data size Performance
  • Scaling search

    Unlike math and coding, search only sees modest gains from increasing model size. Instead, search benefits from looking at more documents. And the cost to see each document increases in lockstep with parameters, limiting the amount of exploration a large model can do.

    SID Large model Pipelines Question difficulty Exploration
  • Scaling swarms

    We start with the atomic unit of search, a small agentic model that reasons and searches in just 300ms and scale it up to dozens of coordinating agents, doing hundreds of searches.

    SID swarm Large model

Built for the most demanding applications.

Performance

2x

more relevant results than embeddings and rerankers

Speed

8x

faster than frontier models

Cost

20x

cheaper than frontier models

Find all

SID-2 is trained to find everything

  • Agent memory
  • Recruiting
  • E-commerce
  • Customer support
  • Healthcare and insurance
  • Legal
  • Finance
  • Procurement and compliance
  • Knowledge bases

Swarms

Scales from super fast single agents to a swarm of 24 for the hardest tasks

SID

4 agents

Infinite context

Effectively infinite context window with swarms of agents, and 256K tokens for a single agent

One agent 256K

Configurable

Match reasoning effort & metric to your application

Low Medium High nDCG Recall Precision F1

Works with anything

Compatible with your existing search backends and tools. No moving data or reindexing required.

  • Reindexing
  • Moving data
BM25 Vector Filters Other SID Your backend One, some or all

Hosting

Self-serve on SID API, serverless with ZDR, SOC 2 and HIPAA, or self-hosted in your VPC on any major cloud

Your VPC SID API Serverless Self-hosted

Scale

89%

quality maintained at 10B items (vs 10k items)

Our research

SID is a research center working on new methods for search.

Open positions Read all

Guest Post Training SID-1 to beat GPT-5 at search with 1k+ QPS RL rollouts