Skip to content
New

DeepSeek V4.1 Flash free for 7 days: 10M tokens a day after a top-up of $10 or more. Until Nov 30.

How it works
aeondave/vec2textAI, agents and MCP

vec2text

⁨vec2text: embedding-inversion library for reconstructing approximate text from sentence embeddings. Use when working with saved embedding tensors, privacy/inversion research, or AI/ML challenge workflows where you need to load a corrector model and invert strings or embeddings directly.⁩

v2026.11.13-21cd5683 installs+3 in 7 days39 Starssynced 10 hours ago

Install

No NotHoly CLI yet? Install it first.

$ curl -fsSL https://notholy.com/install.sh | sh
$ notholy skills install aeondave/vec2text

Writes to ⁦~/.notholy/skills⁩ after checking the signature and the SHA-256.

SignatureSigned, ed25519
hHa85Embtg…kqJxOIAQ==The CLI checks the SHA-256 and the signature before it writes a file.
Package6ce47e2e…412ca165
6ce47e2ec25d…01e0412ca1651 file · ⁦1.1 KB⁩
ScanMedium1 finding
  • No license found
See the full scan
Sourcegithub.com/AeonDave/malskill/tree/main/ai/vec2text@21cd568

Use vec2text when embeddings are the artifact and text recovery is the question.

When to use vec2text

Use vec2text when you need to:

  • invert sentence embeddings back into approximate text
  • load a pretrained corrector for supported embedding families
  • work from saved embedding tensors rather than original source text
  • tune inversion quality with multiple refinement steps or beam search

Quick Start

python
import vec2text

corrector = vec2text.load_pretrained_corrector("gtr-base")

High-Value Workflows

Invert known strings through the matching embedder workflow

python
vec2text.invert_strings(
    ["example text"],
    corrector=corrector,
    num_steps=20,
    sequence_beam_width=4,
)

Invert embeddings directly

python
import torch

embeddings = torch.load("embeddings.pt", map_location="cpu")
texts = vec2text.invert_embeddings(
    embeddings=embeddings,
    corrector=corrector,
    num_steps=20,
)
print(texts)

Practical Notes

  • Corrector choice must match the embedding family; mismatched models produce garbage.
  • num_steps improves refinement quality, while sequence_beam_width improves search at higher memory cost.
  • The project supports both direct string inversion workflows and embedding-only inversion workflows.

Caveats

  • Reconstruction is approximate, not guaranteed exact text recovery.
  • Beam search raises memory usage quickly.
  • Model support and pretrained aliases evolve; prefer current upstream docs over old examples.

Resources

No bundled scripts/, references/, or assets/. Use the upstream vec2text README for current pretrained corrector aliases, API entry points, and embedding-family matching guidance.

Files in this version

Permissions

PermissionDeclaredFound by the scan
Reads filesNoNo
Writes filesNoNo
Runs shell commandsNoNo
Uses the networkNoNo
Uses connectorsNoNo
Drives a browserNoNo

Scan findings

Reviewed on Nov 13, 2026

  1. MediumLicense · No license found

A clean scan lowers the risk but does not prove a package is safe. Read what you install.

VersionPublishedCommitSHA-256SignaturePermission change
v2026.11.13-21cd568Nov 13, 202621cd5686ce47e2eSigned–

Ask anything your scope allows.

Create an account, add credits and ask the first question in the chat or the terminal.

  1. 1/3

    Create the account

    Individual or company. You can switch to a company later without losing balance or keys.

    IndividualCompany
  2. 2/3

    Add credits

    By card. Pay per token, no subscription.

    $5.00minimum deposit
  3. 3/3

    Ask

    In the chat, or install the CLI and run notholy login in your terminal.

    $ notholy login