Atlas

Because "trust me bro" isn't an enterprise retrieval strategy.

Atlas carries the context so your LLM stops hallucinating with 100% confidence.

Pipeline & Workbench Modules 8 Interactive Modes

01

Ingest Feed files

02

Chunk Chop chop

03

Embed Math magic

04

Pipeline Wire DAG

05

Execute Pull trigger

06

Eval Receipts

07

Chat Talk to docs

08

Settings System config

Document Ingestion

Parse and index documents into your local SQLite registry for chunking and retrieval.

Total 0 files

Ingested 0 ready

Pending 0 waiting

Progress 0% Ingested

Select any file from the sidebar to inspect or ingest individually.

Chunking Laboratory

Active File: —

0 chunks

Strategy

Parameters

chunk_size

overlap

Actions

Chunks Preview

Select a strategy and generate chunks

Chunking Strategies

Partition parsed text into optimal segments for vector embedding.

Total

0

Chunked

0

Pending

0

1. Recursive Character

Splits text progressively by paragraphs, newlines, and spaces with overlap.

Use Case: Articles, Markdown notes, PDFs, general documentation.

2. Code-Aware

Splits source code along functions, classes, and language syntax boundaries.

Use Case: Python, JavaScript, TypeScript, and software codebases.

3. Semantic

Calculates embedding similarity and splits when there is a shift in topic.

Use Case: Research papers, essays, long-form narratives.

4. Markdown Structure

Preserves header hierarchy and splits content cleanly by document sections.

Use Case: Technical spec sheets, user guides, API docs, README files.

5. Sentence Window

Extracts single sentences and attaches surrounding context windows.

Use Case: Fact retrieval, precision Q&A, clause lookup.

6. Fixed Size

Splits text into fixed character or token lengths with uniform overlap.

Use Case: Baseline benchmarks, simple uniform text files.

Select any document from the sidebar to inspect or test chunks

Embedding Studio

Vector generation for —

Embedding Model

BAAI/bge-small-en-v1.5

384-dim · Local Fast

all-MiniLM-L6-v2

384-dim · Balanced

all-MiniLM-L12-v2

384-dim · Quality

intfloat/e5-small-v2

384-dim · E5 Family

paraphrase-MiniLM-L3

384-dim · Ultra Fast

Embed Strategy

Embedding Cache

Ready to Index

Vector Space Projection

Vectorize document first, then click Project

Semantic Neighbor Search

Search nearest neighbors in vector space

Embedding Models

Transform partitioned chunks into dense semantic vector representations.

Total

0

Embedded

0

Pending

0

1. BAAI BGE Small

384-dim

BAAI/bge-small-en-v1.5

High-efficiency dense retrieval model optimized for rapid vector indexing and query ranking.

Use Case: Real-time local RAG, fast document search, quick benchmarks.

2. All-MiniLM-L6

384-dim

all-MiniLM-L6-v2

Fast, lightweight general-purpose sentence transformer with minimal memory and CPU execution.

Use Case: Everyday semantic embeddings, query similarity, lightweight CPU deployment.

3. All-MiniLM-L12

384-dim

all-MiniLM-L12-v2

Deeper 12-layer transformer providing superior sentence clustering and nuanced representations.

Use Case: High-accuracy document clustering, nuance-heavy retrieval, semantic deduplication.

4. Intfloat E5 Small

384-dim

intfloat/e5-small-v2

Asymmetric retrieval model trained specifically on query-to-passage search matching.

Use Case: Knowledge base Q&A, customer support search, query-to-passage matching.

5. Paraphrase MiniLM-L3

384-dim

paraphrase-MiniLM-L3-v2

Ultra-compact 3-layer distilled transformer designed for instant inference and semantic equivalence.

Use Case: Real-time duplicate detection, paraphrasing, intent matching, low-latency APIs.

Select any document from the sidebar to vectorize chunks or project 2D UMAP embeddings

Execution Pipelines

Select a pipeline to inspect its composition, architecture and target use case.

8 Presets

Pipelines 8

Pipeline

TOP-K

MODEL

1. Doc Ingestion Parsed & indexed

CHECKING

2. Chunker Strategy Partitioned strategy

CHECKING

3. Vector Embeddings Target strategy vectors

CHECKING

Ready to Execute Pipeline

Select a pipeline above, enter a prompt, and click Run to retrieve grounded context and generate a response.

Execution Runs

0 runs

Execute Tab Runs

Loading recent runs...

Chat Tab Runs

No recent chat runs found.

Retrieved Chunks

0 chunks

Showing retrieved chunks

Loading all retrieved chunks...

RAG Arena

Head-to-head multi-pipeline performance combat.

Triad Radar Footprint

Metric Breakdown

Arena Leaderboard

Awaiting Ignition... Click "IGNITE ARENA" to evaluate all selected pipelines.

New Conversation

Pipeline

TOP-K

MODEL

1. Ingest Parsed

CHECKING

2. Chunk Partitioned

CHECKING

3. Embed Vectors

CHECKING

Document Intelligence Assistant

Ask questions, synthesize insights, or query complex data grounded directly in your indexed documents.

Conversations

History 0

No sessions yet

Atlas

Document Ingestion

Chunking Laboratory

Chunking Strategies

1. Recursive Character

2. Code-Aware

3. Semantic

4. Markdown Structure

5. Sentence Window

6. Fixed Size

Embedding Studio

Embedding Models

1. BAAI BGE Small

2. All-MiniLM-L6

3. All-MiniLM-L12

4. Intfloat E5 Small

5. Paraphrase MiniLM-L3

Execution Pipelines

Create Custom Pipeline

Ready to Execute Pipeline

Execution Runs

Execute Tab Runs

Chat Tab Runs

Retrieved Chunks

RAG Arena

Triad Radar Footprint

Metric Breakdown

Arena Leaderboard

New Conversation

Document Intelligence Assistant

System Configuration

Project Status

Active LLM Provider

API Keys & Endpoints