EmbedFlow

Progressive embedding-model migration over existing vector indexes.

EmbedFlow lets a new embedding model serve over candidates from an existing vector index while target document vectors are materialized progressively. It supports migration analysis, persistent caching, background work, and serving through FAISS, Qdrant, a CLI, and FastAPI.

Quickstart · Documentation · Research

Why EmbedFlow?

Embedding-model upgrades usually mean re-embedding the corpus and building a second index before the new model can serve. EmbedFlow tests whether the existing retriever can remain useful during that transition.

The core observation is simple:

Different representation spaces can still preserve useful retrieval neighborhoods.

flowchart LR
  Q[Query] --> S[Source model]
  S --> I[Existing index]
  I --> C[Top-K candidates]
  C --> T[Target scoring]
  T --> R[Results]
  C --> M[Materialization queue]
  M --> V[(Target vector cache)]
  V --> T

Measured candidate gap from the registry:

Install

Install the published package from PyPI:

python -m pip install embedflow

For FAISS and the dashboard, add the optional integrations:

python -m pip install "embedflow[faiss,dashboard]"

Qdrant and model-runtime extras are documented in docs/installation.md. For model-backed analysis, install embedflow[faiss,models,dashboard].

Try it

The deterministic demo needs no paid service or model download: install the FAISS/dashboard variant above to use its browser UI.

embedflow demo

Open http://127.0.0.1:8000/. The first search can be COLD or PARTIAL; repeated traffic becomes WARM as the background materializer fills the persistent cache.

For a setup-only run, pass --no-serve:

embedflow demo --no-serve

The repository also includes scripts/run_demo.sh for source-checkout development.

Analyze a migration

For an existing source index and no native target index, run the finite-tail analysis first:

embedflow analyze \
  --documents ./documents.jsonl \
  --index ./legacy.index \
  --source-model sentence-transformers/all-MiniLM-L6-v2 \
  --target-model Qwen/Qwen3-Embedding-0.6B \
  --probe-queries ./probe_queries.jsonl \
  --model-root ./models \
  --device cuda \
  --output-dir ./analysis

The report gives a T2-v1 diagnostic (SAFE, EXPAND, or UNSAFE_OR_UNCERTAIN), a recommended initial candidate depth, and ANN health. Treat SAFE as an empirical deployment signal and validate important migrations on the target corpus. Use --device cpu on a CPU-only machine.

Start progressive serving with the generated configuration:

embedflow serve --config ./analysis/embedflow.analysis.yaml --device cuda

The Python facade is available when an application needs an in-process session:

import embedflow

session = embedflow.migrate(
    index="./legacy.index",
    old_model="sentence-transformers/all-MiniLM-L6-v2",
    new_model="Qwen/Qwen3-Embedding-0.6B",
    documents="./documents.jsonl",
    model_root="./models",
    device="cuda",
    candidate_depth=50,
    cache_path="./embedflow_cache",
)
results = session.search("what causes auroras?", top_k=10)

Search responses expose COLD, PARTIAL, or WARM, cache hits and misses, synchronous work, queued work, and stage timings. Once the candidate vectors are warm, target scoring over that candidate set is deterministic.

Known migration evidence

EmbedFlow ships a versioned core registry of measured results from the research study. Matching model contracts can provide useful starting depths and show what was observed on earlier corpora; a new corpus still receives its own analysis.

embedflow registry list
embedflow registry show \
  --source Qwen/Qwen3-Embedding-4B \
  --target Qwen/Qwen3-Embedding-8B
embedflow registry match --config ./embedflow.yaml

Selected core records (nDCG@10, G(50)):

Exact corpus and contract matches can reuse canonical results with --use-registry. Matching contracts on a different corpus are reported as prior evidence and still trigger current-corpus validation. See docs/registry.md for matching and provenance details.

Research

For candidate depth K, EmbedFlow measures:

G(K) = M_T - M_{T|S_K}

Lower G(K) means the source candidate pool recovers more of native target retrieval quality. Containment reports neighborhood overlap separately. When native target evidence is available, the observed depth is K*_epsilon = min { K : G(K) <= epsilon }. With probe data alone, T2-v1 estimates finite-tail behavior and recommends an initial depth.

The study includes 63 development settings, frozen BRIGHT validation, and Natural Questions scale experiments through 1M documents. T2-v1 is the frozen finite-pool diagnostic used before a native target index exists. ANN fidelity is measured separately and is UNKNOWN until an exact reference is supplied.

• Concepts
• Methodology
• Known evidence registry
• Paper: EmbedFlow: Upgrading Legacy Embeddings Without Full Upfront Re-Embedding

Integrations

Backend-specific setup and examples:

• FAISS
• Qdrant
• Adding a backend

CLI

embedflow --help
embedflow analyze --help
embedflow serve --config ./embedflow.yaml
embedflow status --config ./embedflow.yaml
embedflow registry list
embedflow economics --corpus-size 1000000000 --docs-per-second 106.98 --gpu-price 3.29
embedflow doctor --config ./embedflow.yaml

The full command reference is in docs/cli.md. The FastAPI service exposes health, status, search, analysis, prewarming, metrics, and OpenAPI documentation; see docs/api.md.

Documentation

• Installation and extras
• Quickstart
• Configuration
• CLI reference
• API
• Economics
• Limitations
• Contributing
• Security

Status

EmbedFlow v0.1.0 is an alpha release for research and early real-world testing.

• T2-v1 reports an empirical finite-tail diagnostic.
• PARTIAL rankings can differ from fully warm target reranking.
• ANN fidelity needs a reference comparison to audit.

Citation

The accompanying paper is EmbedFlow: Upgrading Legacy Embeddings Without Full Upfront Re-Embedding. The public paper URL is coming soon. Citation metadata is in CITATION.cff.

License

Apache-2.0. Copyright 2026 Arnav Srivastav.