annpack 0.1.7

ANNPack builder and tools

ANNPack

Serverless vector search: static .annpack files served over HTTP Range, searched in the browser via WASM + Transformers.js.

Badges

TL;DR / What it is

• A portable, immutable ANN index format plus tooling for build + serve + search.
• Designed for low-ops distribution (CDN/S3/edge) and in-browser query.
• Ships a Python CLI/API and a WASM runtime for HTTP Range search.

Alpha / security posture: ANNPack is in alpha. Security hardening is in progress; see SECURITY.md for current defaults, limits, and runtime caps.

v0.1.5 highlights:

• Deterministic tiny-dataset fallback for centroid training and pinned FAISS threads to stabilize builds.
• Hardened registry defaults (dev mode opt-in, JWT required in prod), upload/body limits, rate limiting, and path traversal protection.
• C core overflow/alloc guards plus Python metadata size caps (load_meta=False support), with new security workflows/docs.

When to use / When not to use

Use it when:

• You want static, cacheable search artifacts you can host anywhere.
• You need reproducible snapshots and lightweight, client-side querying.

Avoid it when:

• You need mutable, transactional, or real-time index updates.
• You want a managed vector database with server-side filtering and writes.

Quickstart (CLI)

pip install annpack
ANNPACK_OFFLINE=1 annpack build --input tiny_docs.csv --text-col text --output ./out/tiny --lists 256
annpack serve ./out/tiny --port 8000
annpack smoke ./out/tiny --port 8000

Expected output includes PASS smoke.

Installation

pip install annpack

Optional extras:

• pip install annpack[embed] for real embeddings (torch + sentence-transformers)
• pip install annpack[registry] for the local PackHub service

Templates (plug-and-play)

annpack templates
annpack init --template text-search --out ./my-pack

Concepts

• A pack is an immutable index directory containing:
  • *.annpack (binary index)
  • *.meta.jsonl (metadata rows)
  • *.manifest.json (shard list; used by the UI)
• PackSet supports base + delta packs with tombstones for append-only updates.
• Components:
  • Builder (Python CLI): annpack build
  • Runtime (C/WASM): ann_load_index, ann_search over HTTP Range
  • Frontend (JS): Transformers.js embeddings + WASM ANN search

Offline mode & determinism

Set ANNPACK_OFFLINE=1 to use deterministic dummy embeddings (no model downloads). For real embeddings, install annpack[embed] and unset ANNPACK_OFFLINE.

Determinism notes:

• Manifests/meta are deterministic and clustering is seeded.
• Embeddings can vary by device/backend; for strict reproducibility, set ANNPACK_DEVICE=cpu during builds.

Demos

10-minute demo (offline, deterministic):

bash examples/quickstart_10min.sh

Expected output includes:

• PASS smoke
• READY: open http://127.0.0.1:<port>/

Golden demo launcher (build + serve + UI command):

bash tools/run_demo.sh

Optional medium demo assets (downloaded, not in git):

bash tools/download_demo_assets.sh ./demo_assets

Troubleshooting

• Port in use: pass --port <free-port> to serve/smoke.
• Missing manifest: ensure the output dir contains *.manifest.json.
• CORS: annpack serve enables permissive CORS headers by default.
• Offline/air-gapped builds: set ANNPACK_OFFLINE=1 (no embed deps required).
• Real embeddings: install annpack[embed] and unset ANNPACK_OFFLINE.
• Small datasets: annpack build clamps --lists to vector count.
• macOS quarantine (console scripts):

  xattr -dr com.apple.quarantine "$(python -c 'import site; print(site.getsitepackages()[0])')"
  

• Localhost bind restrictions: set ANNPACK_SKIP_NET_TESTS=1 to skip network smoke in stage_all.sh.
• Avoid venvs named # (shell treats # as a comment).

Verification / acceptance scripts

Stage 1 acceptance (build + smoke):

bash tools/stage1_acceptance.sh

Expected last line: PASS stage1 acceptance.

Stage 2 acceptance (API + determinism):

bash tools/stage2_acceptance.sh

Expected last line: PASS stage2 acceptance.

Stage 3: Delta packs (PackSet):

annpack packset build-base --input tiny_docs.csv --packset ./packset --text-col text --id-col id --lists 4 --seed 123 --offline
annpack packset build-delta --base ./packset/base --add delta_add.csv --delete-ids 1 --out ./packset/deltas/0001.delta --packset ./packset --text-col text --id-col id --lists 4 --seed 123 --offline
annpack packset status --packset ./packset

Stage 4 acceptance (distribution + portability):

bash tools/stage4_acceptance.sh

Expected last line: PASS stage4 acceptance.

Pre-talk gates (clean repo + clean checkout):

bash tools/repo_hygiene.sh
bash tools/clean_checkout_gate.sh

Repo layout

• python/annpack/: primary Python package and CLI.
• registry/: local FastAPI-based pack registry with Range support (registry/README.md).
• web/: client SDK (web/packages/client) and UI app (web/apps/ui).
• docs/: architecture, API, CLI usage, and notes.
• annpack-v2/: experimental WASM demo tooling; treat as legacy.
• Legacy notes:
  • build_fast.py is legacy; use annpack build with MiniLM instead.
  • ann_query_bytes in main.c is retained for debugging.

Docs

• docs/ARCHITECTURE.md
• docs/API_USAGE.md
• docs/CLI_USAGE.md
• docs/WASM.md
• docs/POSITIONING.md
• docs/AUDIENCE_FRONTEND.md
• docs/AUDIENCE_ML_INFRA.md
• docs/VERIFY.md
• docs/BENCHMARKS.md
• docs/benchmarks/README.md
• CODE_OF_CONDUCT.md
• SUPPORT.md
• SECURITY.md
• RELEASE.md
• docs/PRODUCTION.md
• docs/COMPATIBILITY.md
• docs/ONE_PAGER.md
• docs/DISCOVERY_QUESTIONS.md
• docs/DEMO_SCRIPT.md

File format

ANNPack is a portable binary format for IVF-based ANN search. The current spec is in docs/FORMAT.md.

Format summary (unchanged):

• 72-byte header: magic, version, endian, header_size, dim, metric, n_lists, n_vectors, offset_table_ptr.
• Centroids (float32), then IVF lists: [count:u32][ids:int64*count][vecs:float16*dim*count], then an offset table [offset:u64,length:u64] per list.

Pure Python reader/searcher example:

import numpy as np
from annpack.reader import ANNPackIndex

with ANNPackIndex.open("my_index.annpack") as idx:
    dim = idx.header.dim
    q = np.random.randn(dim).astype(np.float32)
    q /= np.linalg.norm(q)
    hits = idx.search(q, k=10)
    print(hits)

Other languages can implement a reader using docs/FORMAT.md; the C/WASM runtime is one consumer.

License / Contributing / Security

• License: LICENSE
• Contributing: CONTRIBUTING.md
• Security: SECURITY.md