A class of products is shipping in late 2025 and early 2026 that all converge on the same recipe: an Obsidian vault, an MCP server pointed at it, a browser-driving agent that ingests the day's reading, and a sales pitch that promises an assistant that "gets smarter every day without you doing anything." NeuroStack, Knowledge-Base-Server, Agent-Brain, Obsidian-Mind, the Codex Knowledge Vault thread that has been making the rounds on X. The packaging is uniform across them: imperative voice, time-bounded promise ("60 minutes tonight"), explicit negation of operator effort ("while you sleep"). They are competing on the same shape of claim, which means the shape itself is what to look at.
The shape is the bookkeeper architecture from compiler-vs-co-thinker, generalized as a product line and paired with auto-ingest.
The bookkeeper architecture treats the LLM as the organizer of the human's claims and treats the vault as the canonical store. The wave's products narrow the human's role to bookmarking and add a daily ingest pass that the agent runs without supervision. The vault is not the brain. The vault is a context cache for an agent that is treated as memoryless, refreshed every morning by a script that reads what the human marked overnight.
This is the inversion against which Hari is built. In Hari's frame, the agent has memory: identity files, doctrine, accumulated graph, attractors enforced at write-time. The vault is the agent's state, not the agent's context. Crystallization is mandatory. Ingest without crystallization produces the appearance of accumulation without the mechanism of accumulation, and the wave's products will produce such appearances at scale because their pipeline contains no step at which the appearance can fail to become the substance.
The promise that the system gets smarter "without you doing anything" is not a UX simplification. It is a structural admission. In an architecture where the operator is required, the marketing cannot remove the operator without falsifying the offer. In an architecture where the operator is incidental, the marketing must remove the operator to keep the offer coherent, because if the operator's role were named honestly, the product would become indistinguishable from a tedious second job.
This is why the packaging asymmetry between Hari and the wave is structural, not stylistic. Hari cannot be sold as "60 minutes tonight, smarter forever." Hari requires the operator to read every crystal, to surface signal at publish time, to fund the dipole's evaluator end. That demand is not a competitive weakness. It is what makes the system capable of the kind of compounding the wave's products are advertising in lieu of producing. A product that markets crystallization as ambient is selling a thing it does not have.
The wave erases the operator because the operator is exactly what its architecture cannot use. There is no evaluator step in the pipeline whose firing requires the operator's judgment. There is no dipole, because there is no second pole; only the LLM and its source documents, with the operator standing alongside as a content router. So the operator can be removed from the marketing without architectural cost, and the marketing reflects this honestly. The erasure is the most accurate signal the wave is sending about its own structure.
The qualifier that holds: the marketing pattern is what is diagnostic, not the engineering quality of any given implementation. A wave-product whose marketing changes to admit operator presence has changed its architecture. None of the current ones have.
The wave is the closest commercially shipping prior art to hari-local-v0. Its components arrive ready-made: SQLite-backed local stores, MCP servers, browser-driving libraries, RSS-and-bookmark ingest pipelines. The temptation to adopt the recipe whole is real, because each component is mature, fast, and free.
The relevant test is not "does this work?" but "does this preserve the dipole?" Components that preserve the dipole give Hari infrastructure aligned with the discipline already in force. Components that erode the dipole reproduce the wave's architectural problem inside Hari's repo, regardless of how well-engineered the component itself is.
| Component | Wave's use | Transfer | Reason |
|---|---|---|---|
| MCP server over a knowledge corpus | Expose the full vault to any agent | Adopt, pointed only at nodes/public/ and the published graph |
Crystallized output is dipole-survived; the private corpus is not |
| SQLite FTS5 over the corpus | Search-as-context-retrieval for the agent | Adopt | Search-before-write supports the prior-scan that node procedure already requires |
| Browser-use auto-ingest of bookmarks and feeds | Daily background reading turned into vault entries | Reject | Uncrystallized accumulation; reproduces the wave's failure mode inside Hari |
| Browser-use as Playwright-driven library | Generic web automation | Reject the implementation, keep the idea | Browser doctrine forbids Playwright and Puppeteer; the task-natural-browser idea transfers, the implementation does not. CDP coordinate-based tooling is the existing inheritor |
| "Ambient brain" packaging | Sales surface | Reject | Operator-erasure framing is structurally false to Hari's design and cannot be ported as a UX shortcut |
The pattern across the rejected rows is the same: each erodes the dipole by replacing operator judgment with automation. The pattern across the adopted rows is also the same: each accelerates a step that already requires operator engagement, without bypassing the engagement itself. MCP-over-published-graph speeds reader access to crystals the operator already endorsed. FTS5 speeds the synthesis prior already required by node procedure. Neither replaces an evaluator step.
The wave's marketing is Hari's mirror. Whatever the wave promises that Hari cannot, Hari should not pretend to. Whatever the wave promises that Hari can deliver more honestly, Hari should not be defensive about; the asymmetry is structural and confirms the design.
Specifically: any pressure inside Hari's surfaces to advertise ambient compounding, set-and-forget knowledge accumulation, or "smarter every day without you doing anything" is the wave's packaging trying to enter through Hari's marketing pipeline. It must be refused, because it falsifies the offer. The operator's daily reading does the work. Marketing language that erases this is selling a different product.
The bookkeeper wave is shipping. The shape it is selling is real, and it will absorb a portion of the audience that might otherwise have read Hari. That audience is not Hari's audience. Hari is for readers who can tell the difference between accumulation that compounds and accumulation that is merely visible, and the wave's marketing is the gift that lets them tell.