Collapse worker storage and search into HelixDB (beta) - #1
Draft
Gand4lf wants to merge 22 commits into
Draft
Conversation
…hedotmack#3468) Adds a ninth observation type to the code mode: `sensitive` — information that isn't quite private, but that you wouldn't want leaking into further content development in the wrong context (internal URLs, unreleased plans, personal details, business metrics, client or partner names). These fire a Telegram notification by default, alongside security_alert. The notifier already matched obs.type against CLAUDE_MEM_TELEGRAM_TRIGGER_TYPES, so wiring notifications needed only the new type in the default trigger set plus its own emoji. - plugin/modes/code.json: new `sensitive` type. Also fixes pre-existing staleness in type_guidance, which still said "6 options" and never listed the security types from thedotmack#2084 — that comment is the only type prose the observer agent sees, so a type missing from it is invisible to the model. - SettingsDefaultsManager: default trigger types now security_alert,sensitive, plus a migration for the legacy persisted value. A fresh settings.json is seeded with every default and persisted values win on load, so installs created after thedotmack#2084 had "security_alert" frozen on disk and could never receive the new default — the feature would have been dead on arrival for them. Only the exact legacy value is rewritten; customized lists are left alone. (Greptile P1.) - TelegramNotifier: sensitive observations post with their own emoji - bmp-safe: astral fallback for the new emoji (issue thedotmack#2787 failure class) - CorpusRoutes: allow the new type in knowledge-corpus filters - openclaw feed + weekly-digests legend: recognize the new type Parser validation, the observer prompt type enum, context legends, and the viewer all derive from the mode file, so they pick this up with no change. Regression coverage for the migration: legacy-default migration, write-back with unrelated keys preserved, customized list preserved, superset list preserved, empty opt-out preserved, idempotency across repeated loads.
…edotmack#3513) The installer's provider prompt now leads with CMEM Pro and shows what each option actually costs per 1,000 observations, so the choice is made on price rather than on brand recognition. Anthropic moves last — it is the most expensive per observation and it bills the user's own Claude plan. Picking CMEM Pro opens cmem.ai/pro?from=installer, waits for the key the signup flow hands back, and writes it into settings. No new provider class: the worker's OpenRouter client is already a generic OpenAI-compatible client whose base URL and model come from settings, so CMEM Pro is four settings writes (provider=openrouter, base URL, model, key). 'cmem' is a prompt-only sentinel and never reaches settings.json, which the worker would not understand. Cost strings live in the new src/npx-cli/cmem-pro-costs.ts and derive from a single TOKENS_PER_OBSERVATION constant, so re-pricing is a one-line edit. CMEM_PRO_ORIGIN overrides the origin for walking the flow against a dev server. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
…otmack#3515) The provider prompt quoted constants derived from a deck. Model pricing changes constantly and old models tend to get more expensive, so those figures were wrong within weeks of shipping: deepseek/deepseek-v4-flash hardcoded $0.091/M live $0.1418/M +56% anthropic/claude-haiku-4.5 hardcoded $0.297/M live $1.0520/M +254% google/gemini-2.5-flash-lite hardcoded $0.113/M live $0.1039/M -8% Rates now come from OpenRouter's public model catalogue at prompt time, blended at claude-mem's real 98.7/1.3 input-output split. The lookup is bounded to 3s and falls back to last-known figures, so an offline install still gets a working prompt. A partial catalogue response counts as a fallback rather than reporting itself live. TOKENS_PER_OBSERVATION goes from an assumed 30,000 to a measured 55,960 — rounded to 56,000 — taken from 201 real observations. The old value was low by 1.87x, and since every figure derives from it, the two errors compounded. Net effect on what the prompt tells people: OpenRouter $2.73/1k -> $7.94/1k Gemini $3.40/1k -> $5.82/1k Anthropic $8.91/1k -> $58.91/1k The old numbers understated every option that bills the user, which meant the prompt argued against CMEM Pro using CMEM Pro's own figures: at 75M tokens the advertised Anthropic rate worked out to $22.28/mo against CMEM Pro's flat $30, so a reader doing the arithmetic would have concluded Anthropic was cheaper. The live rate puts that same comparison at $78.90. Anthropic's suffix changes from "billed to your Claude plan" to "from your Claude plan". Most people choosing that option are on a Claude subscription and spend quota, not cash; the old wording implied a bill they do not receive. Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
# Conflicts: # plugin/scripts/context-generator.cjs # plugin/scripts/mcp-server.cjs # plugin/scripts/server-service.cjs # plugin/scripts/transcript-watcher.cjs # plugin/scripts/worker-service.cjs # plugin/ui/viewer-bundle.js
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This removes the dual-write, dual-search, and reconciliation machinery. There is now one storage model, one search path, and one source of truth.
Beta status
This is intentionally beta:
Validation