WeChat Intelligence Hub: A Read-Only Local Reader for Turning Chat History into Briefings

What the project actually solves, and for whom

The README frames the goal narrowly: turn a local WeChat history into something searchable, verifiable and actionable. The outputs it names are contact history, group-chat topics, items awaiting reply, promises, business opportunities, reactivation leads and reports for an arbitrary date range. That list is the product. If you do not need those artifacts, the project is not aimed at you.

The intended user is one person working on their own machine. The README describes a personal intelligence library, not a team analytics platform, and every workflow starts from a locally authorized database plus access material. There is no hosted component described, no account system, and no server-side ingestion. The repository ships a read-only CLI, Codex skills, an executable entry point, tests and fully fictional samples.

The README is unusually explicit about what this is not: it says the project is not a prompt bundle, and it says the release is not a case where you install it and it automatically reads everyone's full WeChat history. That second sentence is the most important line in the document, because it contradicts the assumption most people bring to a tool like this.

Four layers, one product: how the reader and the engine are split

The codebase separates WeChat capability into four layers so each can be tested on its own, while presenting a single install to the user. projects/rion-wechat-reader/ holds the clean-room read-only reader core. skills/wechat-cli/ is the unified agent entry point and by default calls only that in-house reader; it calls a compatible backend only when the user explicitly sets RION_WECHAT_CLI_BIN. skills/wechat-intelligence-hub/ carries the agent's invocation entry and judgement rules. projects/wechat-intelligence-hub/ is the deterministic local engine, with fictional samples and tests.

That split matters when you evaluate the project. The reader is the part that touches WeChat data, and the README states it does not fetch keys, re-sign, inject or hook WeChat. The optional experimental access helper has its own authorization and side-effect boundary, documented separately. The engine downstream is deterministic and local, which is why the fictional demo can exercise indexing, judgement and reporting without any database at all.

The public interface of the v0.9.2-preview.2 reader is stated to align with 29 read-only tools and 266 input fields from the older wechat-cli 1.6.19 interface. It can read explicitly supplied schema-2 salt-key authorization material and installs isolated SQLCipher and Zstandard runtime dependencies. Those are the concrete compatibility claims; the README also says other WeChat versions are covered item by item against a capability matrix, meaning coverage is accumulated rather than guaranteed.

Installing it and running a first real query

The README offers two paths. The first is to hand a prepared Chinese prompt to Codex and let it read the repository, check for an existing install, reuse any working configuration or key, handle missing dependencies and stop for your confirmation before anything with side effects. The second is manual. Start by cloning:

git clone https://github.com/Rion-Wu-tech/wechat-intelligence-hub.git
cd wechat-intelligence-hub

Then install the full product. Passing no skill name installs wechat-cli, wechat-intelligence-hub and the local engine; naming only wechat-intelligence-hub still pulls in wechat-cli, so you never assemble the two components by hand.

./scripts/install.sh --with-sqlcipher

If you want a standalone command-line entry instead of the Codex skills, install just the reader and run its diagnostics. The README says ready means an existing configuration is being reused, ready_to_configure means you should run setup with the same input, and needs_access means access material is missing and you should not re-run setup. It also warns that a top-level ok: true only means the diagnostic finished, so read data.state.

projects/rion-wechat-reader/install.sh --with-sqlcipher
rion-wechat-cli self-test
rion-wechat-cli self-test --require-sqlcipher
rion-wechat-cli access-plan --pretty

Before touching real data, run the fictional demo. It exercises the indexing, judgement and reporting chain with no database, which is the only end-to-end path you can validate without access material.

bash projects/wechat-intelligence-hub/scripts/run_demo.sh

Personalization comes next, and it is what makes the briefing rank things for you rather than generically. The README says profile-init turns your current work, background and relationship labels into the sorting basis for the daily report; if you have no personal or plan documents, running it alone produces a local preparation checklist, and reports still generate but are marked as not yet personalized.

cd projects/wechat-intelligence-hub
python3 wechat_intelligence_hub.py profile-init \
  --owner-alias "你的微信昵称" \
  --personal-doc "/path/to/个人说明.md" \
  --plan-doc "/path/to/本月计划.md" \
  --priority-label "你的重点联系人标签"

After reopening Codex, the README's example invocation is a natural-language request for the last 24 hours of WeChat activity. The full report keeps two versions by default: wechat_daily_full.md as the Markdown entry point, a wechat-report/ directory of section Markdown, and wechat_daily_report.html as the interactive report. The HTML side supports global search, section navigation, group-chat filtering and expansion, original link jumps, light and dark themes, printing and per-section Markdown download. Single-object questions and reply drafts stay in Codex by default; you have to ask explicitly for both Markdown and HTML if you want a targeted investigation saved.

Where it breaks: access material, coverage and macOS onboarding

The hardest constraint is stated plainly in the README: a successful install does not mean chats are readable. Full history and live database reading require authorized local database and access material, and even then only cover data already synced to the machine. Without a key, the documented five-step first-access workflow has Codex prepare fixed-version tools and attempt acquisition after review and explicit confirmation, which may restart WeChat and re-sign a copy. The repository does not bundle acquisition tooling, and neither installation nor the daily report triggers acquisition.

The README goes further: the macOS acquisition path on a new machine has not been tested, and availability across WeChat versions is not guaranteed. That is a real adoption risk rather than a footnote, because it means the first-run experience on a fresh Mac is unverified by the maintainers. Version drift is a second risk. The README says reader database compatibility may change with WeChat versions, and that notification previews are only an inbound, non-complete fallback source. A degraded mode exists, but it is not the full history the product is named for.

If you have no database access at all, the fictional demo still runs and the report chain still validates. That is useful for evaluation and useless for actual intelligence work. The project is the wrong tool if you need to monitor someone else's account, if you cannot supply your own authorization material, or if you want a managed service that handles access for you. The README also tells users not to send keys, passwords or databases to maintainers, communities or issues, which tells you where support stops.

How it differs from a general chat-export or LLM summarizer

The closest common alternative is exporting a chat archive and feeding it to a general-purpose model or a note-taking tool with a summarization prompt. The difference in approach is architectural. An export-and-prompt workflow is stateless per request: you paste or upload, the model summarizes, and nothing persists. WeChat Intelligence Hub keeps a local deterministic engine with its own index, then layers agent skills on top, so a query like the last seven days of discussion about a topic resolves against the index and merges context across conversations before summarizing.

The second difference is the boundary around the data source. A general summarizer accepts whatever you give it and has no opinion about where it came from. This project ships a read-only reader that the README says does not fetch keys, re-sign, inject or hook WeChat, plus an optional access helper with a separate authorization boundary. That design costs you setup effort and buys a narrower blast radius.

The third difference is output shape. A summarizer returns prose. This project produces a Markdown entry point, a directory of section Markdown and an interactive HTML report with four entry points: consolidated actions, group-chat briefing, key contacts and a business-signal radar. If you only ever want a paragraph of summary, the extra machinery is overhead. If you want a browsable, filterable artifact you can archive and re-feed to a model, the shape is the point.

Licence, maintenance and what an upgrade costs you

The repository is AGPL-3.0, with a separate COMMERCIAL-LICENSE.md at the top level. The README also describes paid community membership, one-to-one guidance, enterprise training, custom skill and agent work, and brand collaboration, and states that community fees are not a software purchase fee and do not remove technical access limits. If you plan to embed this in a network-facing service, the AGPL and the commercial licence file are the two documents to read before writing code against it; that is a licensing question for your own counsel, not something this article can settle.

On maintenance, the last push to the default branch was on 2026-09-11, and the repository is not archived. The most recent release listed is v0.9.2-preview.2 from 2026-09-04, described as a public preview. The version string itself tells you the stability posture: this is a preview line, and the README describes the WeChat-related code as ready for public testing rather than finished. There is no documented deprecation or migration policy in the README, and the README does not document rollback, so an upgrade that breaks database compatibility leaves you with the troubleshooting prompts in docs/USAGE.md rather than a stated revert procedure.

Upgrade cost is therefore dominated by the reader, not the engine. The deterministic engine and its fictional demo are insulated from WeChat changes; the compatibility surface is the reader's database handling and the capability matrix that the README says is filled in version by version. Budget for re-running the self-test and access-plan after any WeChat update, and treat a green install as unrelated to whether reading still works.

Editorial conclusion

Adopt it if you already hold authorized local access material for your own WeChat database and want deterministic, local briefings rather than a hosted analytics service. Do not adopt it if you expect install-then-read, if you cannot supply schema-2 salt-key material or a local database, or if you need a supported path on a new macOS machine, which the README says is untested. Before relying on it, run rion-wechat-cli access-plan --pretty and read data.state rather than the top-level ok flag, then run the all-fictional demo to confirm the indexing and report chain works before touching real data.

Frequently asked questions

Official sources

1. Issues
2. License: AGPL-3.0
3. README
4. Releases
5. Rion-Wu-tech/wechat-intelligence-hub on GitHub