Cimmich

What the company appears to be: Cimmich is an unofficial, open-source, local-first companion tool for Immich, a self-hosted photo library platform. The author describes it as a memory and intelligence layer that adds identity evidence separation (Face, Head, Body, Presence) to enhance photo organization without replacing or modifying Immich directly.

What changed: This project emerged from the author's personal need to improve photo organization in their own archive (~600 GB over 15 years), evolving from early experiments and a rough prototype into a structured product during OpenAI Build Week using Codex and GPT-5.6 Sol.

Single most important open question: Is there sufficient evidence of real-world adoption or traction to suggest that Cimmich has a viable path to commercialization, or is it purely an experimental contribution to the Immich ecosystem?

Note: All claims are self-reported by the author and unverified. No revenue, customer data, or traction metrics are provided.

The description states that Cimmich is:

• An unofficial, open-source, local-first companion for Immich.
• A separately stored memory and intelligence layer.
• Not designed to replace or write directly to Immich’s database or source media.
• Built using technologies including Docker, Node.js, Svelte, TypeScript, PostgreSQL with pgvector, FFmpeg, OpenCV, GitHub, Codex, GPT-5.6 Sol, and others.

It implements a system separating four types of identity evidence:

• Face – visible face supporting matching
• Head – visible head without usable face evidence
• Body – someone located through body without contaminating face scoring
• Presence – owner knows someone was there, without inventing biometric evidence

Cimmich connects these elements into a local memory graph across people and other archive content (Pets, Places, Things, Events, Trips, Activities, Life periods), enabling Smart Search.

It supports standard, personal, and password-gated private viewing modes. Enhanced matching is optional and off by default, controlled by the owner.

Optional local evidence providers can contribute various observations (face, body, pose, OCR, object, vision). Guided access provides neutral HTTP/JSON access for connecting models or frontier harnesses like Codex, with operator-controlled permissions.

Inference: The product is described as a service that operates independently of Immich's core database but integrates with it via metadata and user-defined connections. It uses AI tools (Codex, GPT-5.6 Sol) in its development process.

The author states:

• Cimmich began with the question: "can a private photo library preserve the complete human memory without pretending every observation is a face match?"
• It aims to improve upon conventional face recognition by handling messy real-world memories where faces are obscured or not clearly visible.
• The tool does not aim to replace Immich but rather enhance it as a companion.

Positioning evolution:

• Started as personal experimentation around photo organization challenges.
• Evolved into a structured product during OpenAI Build Week using AI-assisted development tools (Codex, GPT-5.6 Sol).
• Now positioned as an open-source contribution to the Immich ecosystem.

Claim: Cimmich is intended to be a non-commercial enhancement to Immich, not a competing platform.

Inference: The positioning suggests a niche focus on privacy-conscious users who want more nuanced identity matching than traditional systems offer.

The description states:

• Cimmich targets owners of large, messy photo archives (e.g., 600 GB over 15 years).
• It appeals to users who value privacy and self-hosting.
• Users likely already use Immich and want better organization features without compromising data sovereignty.

ICP inference:

• Self-hosted photo library users with significant media collections (~600 GB+)
• Privacy-conscious individuals or teams managing personal or sensitive content
• Developers or power users interested in extending Immich functionality

Not evidenced: No explicit customer segmentation, personas, or usage patterns beyond the author’s own experience.

The description states:

• Cimmich is open-source under AGPL-3.0-only license.
• It is not affiliated with or endorsed by Immich or OpenAI.
• The project is presented as a contribution to the Immich ecosystem, not a commercial offering.
• No pricing information, monetization strategy, or revenue model is mentioned.

Not evidenced: No business model, pricing structure, or monetization plans are provided. The tool is described as non-commercial and open-source.

The description states:

• Built using Codex powered by GPT-5.6 Sol during OpenAI Build Week.
• Uses Docker, Node.js, Svelte, TypeScript, PostgreSQL with pgvector, FFmpeg, OpenCV, GitHub.
• Implements a cross-task messaging system for coordination between Backend, UI/UX, and Controller tasks.
• Final release candidate targets Immich 3.0.3.
• Stores data separately in PostgreSQL 17 with pgvector.
• Has its own credentials, migrations, backup and recovery boundary.
• Passes 595 service tests and 768 web tests.
• Includes installation and recovery tools.
• Public demo contains 57 synthetic images (51 Cedar House scenes + 6-image Space Trip extension).
• Demonstrates product behavior, not biometric accuracy or fairness.

Inference: The tool uses AI-assisted development methods and integrates with existing open-source ecosystems. It emphasizes local-first architecture and data separation.

The description states:

• The author has been working on the problem for several months prior to Build Week.
• Early work included research, model experiments, semantic search, identity matching, and basic UI shell.
• During Build Week, it was transformed into a full product with testing, installation tools, and synthetic demo.
• A dated Build Week ledger is included in the repository.

Not evidenced: No user base, adoption rate, or performance metrics beyond internal test results. No evidence of real-world usage or feedback from users.

The description states:

• Cimmich is an unofficial companion to Immich.
• It does not compete with Immich but enhances it.
• The author notes that existing tools were limited in addressing messy photo archives (crowded photos, poor lighting, hidden faces).
• It builds on the idea of local-first solutions and privacy-focused organization.

Not evidenced: No direct competitors or market analysis provided. No mention of similar tools or platforms in the space.

Key risks identified:

1. Lack of traction or adoption – No evidence of real-world usage or user feedback.
2. Dependency on AI tooling (Codex, GPT-5.6 Sol) – Reliance on proprietary AI systems may pose long-term sustainability issues.
3. Unproven scalability – The tool is described as a prototype built in a week; no indication of scalability beyond small-scale use cases.
4. No commercial viability – Not positioned for monetization or growth, limiting potential investment appeal.
5. Limited testing scope – Public demo uses synthetic data only, not real-world biometric accuracy.

Inference: The project appears experimental and community-driven rather than commercially viable or scalable.

1. What specific problems in your own photo archive drove the creation of Cimmich?
2. How do you plan to handle feedback from Immich users or developers?
3. Are there any plans for monetization, partnerships, or commercial use beyond open-source contribution?
4. What are the technical limitations of the current implementation that might prevent broader adoption?
5. Can you provide details on how the cross-task messaging system works and whether it can be scaled?
6. How do you intend to maintain compatibility with future versions of Immich?
7. Has there been any interest from the Immich team or community in adopting parts of this work natively?