OmniGet vs Hindsight
OmniGet is a free, open-source desktop app whose AI agents keep a local memory inspired by Hindsight, the agent memory system by Vectorize.
Written by Antonio Leandro, who makes OmniGet. Hindsight is described from its own repository and releases.
Short answer
With 20 memories per answer the two tie, 67.1% for OmniGet and 64.5% for Hindsight. With about 80 memories Hindsight wins, 71.1% against 63.2%. OmniGet's memory needs no server, searches in many languages and uses about 316 MB; Hindsight runs as a server for many users, has client libraries and has published results on the full benchmarks.
Pros and cons, side by side
Written from our study of each app's repository and releases. Every row names the better app, including when it is not OmniGet.
| What matters | OmniGet | Hindsight |
|---|---|---|
| Right answers, 20 memories per answer | 67.1% of 76 LoCoMo questions | 64.5%; a two-question gap, within the noise |
| Right answers, wide context | 63.2% with 80 memories | 71.1% with its default recall, about 82 memories |
| What you run | Part of the desktop app: one SQLite file and two local models | A server with PostgreSQL and pgvector, or an embedded Python server |
| Search in other languages | Multilingual vectors (multilingual-e5-small) and a multilingual reranker | English embedding (bge-small-en) and English reranker (ms-marco) |
| Many users and other apps | One person's memory, shared with the agents on that computer and over MCP | Memory banks on a shared server, with Python, Node.js and Go clients and a CLI |
| Proof so far | This page: one LoCoMo conversation, not released yet | Published LongMemEval and LoCoMo results, reproduced by outside researchers |
| Disk used | About 316 MB | About 2.4 GB with its Python environment and models |
OmniGet wins 3, Hindsight wins 3, 1 tied
Speed is not timed yet: recall time and the time to read a conversation are being measured on the same computer, with the same model and the same concurrency for both apps. Until then, the table above compares results and setup, not speed.
What each app can do
| OmniGet | Hindsight | |
|---|---|---|
| Runs without a separate server or database | Yes | Partly |
| Multilingual search models | Yes | No |
| Answers questions from memory (reflect) | Yes | Yes |
| Summaries built from many memories | Partly | Yes |
| MCP tools for coding agents | Yes | Yes |
| Shared memory server for many users | No | Yes |
| Client libraries for other apps | No | Yes |
What each needs to run
OmniGet's memory lives inside the desktop app in one SQLite file, and its search models run on the computer, so recalling a memory calls no AI model. Hindsight is a server with PostgreSQL and pgvector, or an embedded Python server, and it is built to hold memory banks for many users and apps.
Both systems read the same LoCoMo conversation, 19 sessions of chat between two people, and answered the same 76 questions. Claude Haiku, through the Claude Code CLI, extracted the facts, wrote each answer from the memories each system returned and judged it against the gold answer, so the memory system is the only thing that changed.
Results
Seen enough? Get OmniGet for your system.
Windows may show a SmartScreen warning the first time. Here is what to do.
OmniGet is a desktop app
It runs on Windows, macOS and Linux. There is no Android or iOS version. Send this page to your computer:
The apps in this benchmark
-
OmniGet
- License:
- GPL-3.0
- Built with:
- Tauri (Rust + web UI)
- GitHub:
- 14.4k stars · v0.10.1
- Version tested:
- OmniGet 0.10.1 with the agent-memory branch (not released yet) (September 21, 2026)
- Its own extractors, torrents and direct files on top of yt-dlp.
- An MCP server and AI coding agents in the same app.
-
Hindsight
- License:
- MIT
- Built with:
- Python server with PostgreSQL and pgvector (or embedded pg0)
- GitHub:
- 60.5k stars · v1.2.1
- Version tested:
- hindsight-all 0.10.1 (embedded server, pg0) (September 21, 2026)
- Published state-of-the-art results on LongMemEval and LoCoMo, reproduced by outside researchers.
- Mature consolidation into observations and mental models, with clients for Python, Node.js, Go and a CLI.
- Works with 25+ LLM providers and runs as a shared server for many agents and users.
Where each one wins
At the same context size the scores are within the noise of a 76-question run, where one question is worth 1.3 points. OmniGet's small edge on time questions comes from resolving relative dates with a calendar when it extracts facts. Hindsight's lead with a wide prompt comes from keeping more small details per session, which pays off once there is room for them.
What this does not measure
One LoCoMo conversation with a small model is not the full benchmark. Recall speed is not compared yet, and neither are Hindsight's mental models and directives or OmniGet's corrections and forgetting.
Which one should you pick?
- Choose OmniGet if you want memory for the AI agents you already run on your own computer, in any language, with no server to keep up, and you want to see, correct and forget what they remember.
- Choose Hindsight if you are building an agent product, need one memory server for many users and apps, or want the best accuracy when the prompt can hold a lot of context.
Other Hindsight alternatives
Free, open-source apps that do some of what Hindsight does.
- Open-source yt-dlp GUIs compared Every desktop app we track, ordered by GitHub stars, with its license, latest release and who it suits.
- OmniGet vs ezytdl Electron front end for yt-dlp with advanced options. Compared with OmniGet
- OmniGet vs Media Downloader Qt/C++ front end for several download engines, including yt-dlp, gallery-dl and aria2c. Compared with OmniGet
- OmniGet vs SCrawler Windows app that saves media from social sites and user profiles. Compared with OmniGet
- OmniGet vs Videomass Cross-platform FFmpeg front end for batch conversion with presets; it removed its yt-dlp downloader in version 6.1.12. Compared with OmniGet
- OmniGet vs Vividl Windows front end for yt-dlp built with WPF. Compared with OmniGet
- OmniGet vs aria2 Command-line download utility for HTTP, FTP, BitTorrent and Metalink, with many connections per file. Compared with OmniGet
- OmniGet vs you-get Command-line downloader written in Python that scrapes media from web pages. Compared with OmniGet
- OmniGet vs youtube-dl The original command-line video downloader that yt-dlp was forked from. Compared with OmniGet
- OmniGet vs yt-dlp The command-line audio and video downloader that OmniGet runs on. Compared with OmniGet
- OmniGet vs Open Video Downloader Cross-platform yt-dlp front end built with Tauri and Vue, with a very small installer. Compared with OmniGet
- OmniGet vs Parabolic yt-dlp front end with a native interface: WinUI on Windows and GNOME on Linux, published on Flathub. Compared with OmniGet
- OmniGet vs VidBee Electron app built on yt-dlp that also makes local transcripts, AI summaries and RSS subscriptions. Compared with OmniGet
- OmniGet vs ytDownloader Electron yt-dlp app with a clip trimmer and a video compressor, on many app stores. Compared with OmniGet
- TencentDB Agent Memory on GitHub Tencent Cloud's memory server for agent teams: a memory core, a hub and a proxy that several agents share.
- OpenViking on GitHub Volcengine's context database for agents: knowledge, memory and skills kept as a browsable virtual file system with scoped search.
- ai-memory on GitHub Long-term memory shared by many coding agents, so work started in Claude Code can continue in Codex without explaining it again.
Where OmniGet still loses
We publish this on purpose. It is the list we work from to make OmniGet better.
- Wide context: with about 80 memories in the prompt Hindsight answers 71.1% and OmniGet 63.2%; OmniGet needs to keep more small details per session.
- Summaries: Hindsight keeps observations and mental models across many facts; OmniGet keeps one short note per person or thing.
- Other apps: Hindsight has client libraries and a shared server; OmniGet's memory is reached through the app and MCP only.
Trying OmniGet costs nothing
Coming from Hindsight, nothing needs to be moved to try OmniGet: the agents inside the app start remembering from the first conversation, and an agent outside it can reach the same memory through OmniGet's MCP server with the memory permissions. Hindsight stays the better fit for a shared memory service.
Method
Run it yourself
node scripts/bench/agent-memory.mjs rebuilds the numbers from the per-question results in scripts/bench/agent-memory/. To measure again: the app's `cargo test -p omniget-core --lib assist::memory::bench::locomo_eval -- --ignored` (LOCOMO_K=20 and 80) and `python scripts/bench/agent-memory/hindsight_locomo.py` (default and TOPK=20), both on LoCoMo conv-26 with Claude Haiku through the Claude Code CLI. Data licensed CC BY 4.0: reuse it freely with a link to this page.
Sources
Questions
Is OmniGet a Hindsight alternative?
For the memory of your own AI agents on one computer, yes: it recalls with keyword, meaning, links and time search, runs without a server and ties Hindsight at 20 memories per answer. For a shared memory server with client libraries, Hindsight fits better.
Which one is more accurate?
On this test they tie with 20 memories per answer, and Hindsight is ahead with about 80, 71.1% against 63.2%.
Other benchmarks
Try OmniGet 1.0.0
Free and open source, for Windows, macOS and Linux.
Download