THE APPLICATION LAYER
Kompile App
A modular application framework with 40+ swappable modules. Build one application against a unified interface — swap LLM providers, embedding models, vector stores, and data sources without touching your business logic.
Modular by Design
Every capability is a separate module behind a shared interface. Swap any layer by changing a config — your pipeline code, agent logic, and API endpoints stay identical.
40+
Swappable Modules
7
LLM Providers
5
Embedding Backends
4
Vector Stores
The RAG Pipeline
A complete retrieval-augmented generation pipeline with pluggable stages. Every step is independently swappable.
Pull data from configured sources (Confluence, Slack, web, etc.) or load from local files.
Parse documents through format-specific loaders into raw text with metadata.
Split into retrieval-sized passages using your chosen chunking strategy.
Generate vector embeddings through any configured embedding provider.
Store vectors and metadata in your chosen vector store for retrieval.
Query with HyDE, multi-query, expansion, or step-back transformers. Rerank results.
Pass enriched context to any LLM provider for response generation with guardrails.
Knowledge Graphs & Agents
GraphRAG
Extract entities and relationships from documents, build typed knowledge graphs, detect communities with Louvain and PageRank, and answer questions that require multi-hop reasoning. Backed by Neo4j or the built-in native graph.
Knowledge Graphs
LLM-driven or manual graph construction with concept extraction, entity resolution, schema enforcement, and graph compaction. Native TransE and RotatE embeddings for link prediction. Export to 9+ formats.
ReAct Agents
Reason-Act-Observe agents with pluggable Reasoner, Actor, and Observer interfaces. Parallel and sequential action execution, GraphRAG-backed reasoning, and eval-based refinement loops.
Guardrails
Input guards for PII detection, prompt injection, toxicity, and topic filtering. Output guards for hallucination detection, relevancy scoring, and format enforcement. All configurable per-pipeline.
Module Directory
Every layer of the stack is a swappable module. Pick what you need.
LLM Providers
Swap between OpenAI, Anthropic, Gemini, local models, or CLI-based agents. Every provider implements the same interface — your code never changes.
OpenAIAnthropicGoogle GeminiLocal ModelsCLI AgentsEmbedding Models
Embed with OpenAI, run ONNX models locally, use native SameDiff embeddings, Sentence Transformers, or PostgresML — same retrieval code.
OpenAI EmbeddingsLocal ONNX (Anserini)SameDiff NativeSentence TransformersPostgresMLVector Stores
Store and retrieve vectors with embedded Lucene HNSW, PostgreSQL pgvector, Vespa hybrid search, or ChromaDB. One API, four backends.
Lucene HNSWPostgreSQL pgvectorVespaChromaDBDocument Loaders
Load PDFs (with table extraction), Microsoft Office, Excel, email inboxes, audio files, and web pages. Apache Tika covers everything else.
PDF + TablesMicrosoft OfficeExcelEmail (IMAP)Audio (Whisper)Apache TikaWeb PagesData Sources / Crawlers
Crawl your tools — Confluence, Jira, Notion, Slack, Discord, Google Workspace, OneDrive, Reddit — and feed directly into the ingest pipeline.
ConfluenceJiraNotionSlackDiscordGoogle WorkspaceOneDriveRedditText Chunkers
Choose sentence splitting, recursive character chunking, markdown-aware parsing, token-based splits, or table-aware chunking. All pluggable.
SentenceRecursive CharacterMarkdown-AwareToken-BasedTable-AwareCompute Graph Integrations
Orchestrate complex workflows beyond simple chains. Plug in external engines through the same Kompile interface.
Apache Camel
Enterprise integration patterns for routing, transformation, and mediation.
Drools
Business rules engine for decision logic that changes independently of code.
n8n
Visual workflow automation for building complex multi-step pipelines.
Ready to compile your stack?
Get early access and start building on the modular AI platform.
Request Early Access