OneContext
The fastest way to ship enterprise-grade RAG pipelines to millions of users.
Near infinite combinations.
Compose your entire RAG pipeline in one simple yaml.
We provide state-of-the-art "steps" for embedding and retrieval pipelines straight out of the box. Just make a list of the steps you want to use, and you're done.
You concentrate on your actual business, and let OneContext handle the Machine Learning devops.
1index:
2
3name: example_index
4
5steps:
6 - step: Preprocessor
7 name: example_pre_processor
8 step_args:
9 add_punctuation: false
10 remove_whitespace: false
11 depends_on: [ ]
12
13 - step: Chunker
14 name: default_chunker
15 step_args:
16 chunk_size_words: 320
17 chunk_overlap: 30
18 split_by: word
19 split_respect_sentence_boundary: true
20 hard_split_max_chars: 2400
21 depends_on: [ simple_pre_processor ]
22
23 - step: SentenceTransformerEmbedder
24 name: sentence-transformers
25 step_args:
26 model_name: BAAI/bge-base-en-v1.5
27 batch_size: 4
28 include_metadata: [ title, file_name ]
29 depends_on: [ simple_chunker ]
30
31 - step: LexRank
32 name: lexranker_file
33 step_args:
34 scope: file
35 depends_on: [ sentence-transformers ]
36
37 - step: LouvainCommunityDetection
38 name: louvain_file
39 step_args:
40 assign_labels: gpt-3.5-turbo
41 depends_on: [ lexranker_file ]
42
43 - step: HdbScan
44 name: hdbscan
45 step_args:
46 min_cluster_size: 6
47 assign_labels: gpt-3.5-turbo
48 depends_on: [ lexranker_file ]
49
50 - step: UpdateOnDb
51 name: db_updater_step
52 step_args:
53 depends_on: [ louvain_file ]
54
55
56query:
57
58name: query_pipeline_1
59
60steps:
61
62 - step: Retriever
63 name: demo_retriever
64 step_args:
65 query: $RETRIEVER_QUERY
66 model_name: BAAI/bge-base-en-v1.5
67 top_k: $RETRIEVER_TOP_K
68 metadata_json: { }
69 depends_on: [ ]
70
71 - step: LexRank
72 name: lexranker_demo
73 step_args:
74 depends_on: [ demo_retriever ]
75
76 - step: LouvainCommunityDetection
77 name: louvain_demo
78 step_args:
79 depends_on: [ lexranker_demo ]
80
81 - step: FilterInMemory
82 name: demo_filter
83 step_args:
84 key: lexranker_demo.percentile_score
85 comparator: $gt
86 value: $EXTRACT_PERCENTAGE
87 depends_on: [ louvain_demo ]
88
89 - step: Reranker
90 name: oc_reranker_test
91 step_args:
92 query: $RERANKER_QUERY_WILDCARD
93 model_name: BAAI/bge-reranker-base
94 top_k: $RERANKER_TOP_K_WILDCARD
95 depends_on: [ demo_filter ]Create and run custom tests
Evaluating RAG pipelines used to be difficult. Not anymore.
Define custom tests specific to your use-case. Compose and upload custom test-sets for your RAG pipelines to operate on. Tests are automatically fired on each new deployment, so you can instantly isolate and fix issues, before they hit your users.
Go from localhost, to production, blazingly fast.
Ship to millions of users with one line of code.
Deploy your pipeline on a Kubernetes cluster, in your compute environment, with the latest GPUs, and in-built autoscaling, in seconds.
1OneContext.deploy({pipeline: "example_index"})Roll back, forward, and back again.
Compose, evaluate, deploy, repeat.
Pipeline version-control is built into OneContext. Metrics and specifications are all persisted, so you can instantly roll back to any previous version.
Automatic observability.
Observe all of the micro-details from a macro-level.
Pipelines deployed on OneContext's cloud come with automatic metrics and observability.
Don't just take our word for it
OneContext is like the Vercel of RAG pipelines. Super easy iteration and deployment. No other product makes it this easy to deploy ML pipelines to a kubernetes cluster!
Wow, this is fantastic. This makes deployment so much simpler. This is going to save our devops team so much time and frustration.
Integrate directly into your stack.
OneContext integrates with thousands of language models
Check out the quickstart guides in our documentation to get started with your favourite tools and services.
