Vectors of different models are not comparable even at the same size; the
store records its embedder in waggle-memory.json and names the mismatch.
Also drop the weaver references from the cli and server tsconfigs.
- Pinned Ollama 0.32.3 (and rollback 0.32.0) Linux x64/arm64 artifacts with
their published SHA-256 and sizes; .tar.zst archives are decompressed by
Node 22's zstd and unpacked by the system tar.
- install.sh sets ONNXRUNTIME_NODE_INSTALL_CUDA=skip: onnxruntime-node
otherwise downloads CUDA libraries on every Linux x64 install.
- Drop the root sqlite-vec-windows-x64 dependency; the sqlite-vec package
already carries per-platform binaries.
Weaver consolidated and decayed frames of the legacy memory engine; mem0
updates and merges memories itself. The ChatGPT/Claude export import wrote
frames and its screen is hidden in v1 (ADR-0002).
The sidecar already bundles the AWS S3 client through other code, so it
is not a reason to keep whatsapp-web.js external. The TypeScript
compiler, pulled in by Puppeteer's config loader, is.
The WhatsApp card in Settings > Channels now says the client is
unofficial, that WhatsApp may block the account, and that a browser is
downloaded (about 170 MB) when none is installed. While the adapter
looks for or downloads the browser, its status detail is shown above
the QR.
The setup hint told users to send /pair from the "Message yourself"
chat, which the adapter never reads because own messages are dropped.
It now says to send it from another WhatsApp account (UF-030).
Baileys depends on libsignal (GPL-3.0), which the sidecar bundled. The
WhatsApp adapter now drives WhatsApp Web through whatsapp-web.js and
keeps its public shape: start, stop, status with the pairing QR, send
and inbound text. LocalAuth stores the session as a browser profile in
<dataDir>/channels/whatsapp; the old Baileys state is left untouched.
The browser comes from whatsapp-browser.ts: CHROME_PATH or
PUPPETEER_EXECUTABLE_PATH, then an installed Chrome, Edge or Chromium
(Windows, macOS, Linux), then a Chrome for Testing build downloaded on
the first start into <dataDir>/browsers, with progress in the status.
The root .puppeteerrc.cjs stops npm install from downloading a browser.
The sidecar build keeps whatsapp-web.js external: bundled, it fails on
__dirname and would inline LGPL node-webpmux. Text only, so ffmpeg is
never needed.
New dependencies, pinned: whatsapp-web.js 1.34.7 (Apache-2.0) and
@puppeteer/browsers 2.13.0 (Apache-2.0). Transitive: puppeteer,
puppeteer-core, chromium-bidi, webdriver-bidi-protocol (Apache-2.0),
devtools-protocol (BSD-3-Clause), extract-zip (BSD-2-Clause),
fluent-ffmpeg and the rest (MIT or ISC), and node-webpmux
(LGPL-3.0-or-later, shipped as a separate package; needs the license
review). Removed: @whiskeysockets/baileys and libsignal.
UF-020 is now Fixed; ADR-0005 records the consequences.
Add a reference set (100 invented memories across personal memory and two
workspaces, 67 queries: paraphrase, keyword, multi-fact, scope, no-answer)
and a tsx benchmark that loads it into the old engine (MindDB, FrameStore,
HybridSearch, with and without the cross-encoder reranker), the mem0 store
and a plain-cosine baseline, all on Waggle's in-process embedder. It reports
recall@1, recall@5, MRR, latency and misses per engine and query type. The
mem0 half runs on its own (--engines mem0,cosine) once the old engine is
removed. Ollama and the user's own key are optional embedders; the bench
refuses network access to anything else and never reads OPENAI_API_KEY.
Results in docs/oss/research/memory-search-quality.md: mem0 matches the old
engine's chat-recall path (recall@1 0.72, recall@5 0.89 vs 0.90, MRR 0.82
vs 0.83) at a fraction of its latency and is acceptable for v1.
No new dependencies.
- Orchestrator, memory tools, context loader and pattern write-back use a
MemoryStore and a workspace scope instead of MindDB layers; the knowledge
graph tools, cognify, memory linker and feedback handler are gone (graph
hidden in v1, ADR-0002).
- The server opens one mem0 store in <dataDir>/memory/, embedded by the
in-process model on first use, and extracts facts about the user in the
background with the model the user chose (memoryExtraction setting).
- Legacy .mind memories are copied into the store once, in the background;
the legacy files are only read.
- Weaver consolidation, harvest, wiki, memory import, memory lanes, vector
backfill and index reconciliation are removed; agent runs, loops, sub-agent
results, connector fetch and the failed-turn capture write through the
contract. Workspace sessions no longer hold a mind handle.
- FileIndexer and the install audit move to waggle.db; recall date helpers and
the reranker are copied into core.
- The CLI REPL opens the same memory store.
- Analytics are opt-in (decision Q7, ADR-0006). PostHog does not start or
capture anything until the user clicks Allow in a first-run dialog that
says what is sent. Escape or closing stores nothing and asks again. The
existing Privacy & Telemetry toggle in Settings withdraws consent at once.
Builds without VITE_POSTHOG_KEY show no dialog and send nothing.
- Remove the remaining upsell copy: the /team page with its per-seat price,
Stripe and Teams wording in the erase-data dialog, Teams/Enterprise hints.
- Remove team screens (decision Q3): Waggle Dance, team presence in chat,
team storage and sharing in workspace creation, storage and files.
- Hide memory features that are not in v1 (decision Q2) behind one map in
lib/memory-features.ts: the knowledge graph, import history, wiki, agent
memory hooks and Room launches. The map is the seam P1.5 fills from the
memory store's capabilities.
- Say Windows, macOS and Linux instead of "Linux soon".
- Update the e2e specs for the removed and hidden screens.
Also logs UF-029: the Settings "Import Data" button has no handler.
The memory MCP server serves the .mind file directly, so it uses the
substrate's IdentityLayer and AwarenessLayer instead of the waggle.db ones
that @waggle/core now exports. The WaggleDB-only constructors caught the
mismatch at build time.
Waggle kept its own tables in the memory database. Before the memory
engine can be replaced by mem0 they need a home of their own.
- add WaggleDB (<dataDir>/waggle.db). Automations, notifications, held
actions, skill hashes, thumbs feedback, cost tracking, the profile card
and awareness items live there. Their stores take a WaggleDB, so passing
a memory database is a type error.
- the profile card and awareness items are now per scope ('personal' or
'workspace:<id>') in one table each, instead of one table per .mind file
- on the first start, copy these tables from personal.mind and every
workspace.mind (importLegacyMindData). The legacy files are only read,
each scope is copied once and a table only while its target is empty.
- copy the logger, injection scanner, external-memory ingress guard,
WorkspaceManager and the embedders from @waggle/hive-mind-core into
@waggle/core, which now re-exports its own copies
- point the server, agent orchestrator and CLI at waggle.db; update the
tests that seeded these tables in a memory database
ADR-0007 records the layout and the licensing question about the copied
files.
The first full e2e run on this branch failed 19 tests, all on features the
community edition removed: /api/tier, Stripe, /api/compliance, the billing
settings tab, the payment pages and the upgrade dialog. Those checks now
assert that the endpoints and pages are gone (404); tier, billing and
team-settings probes are dropped from the remaining journeys. The rerun is
410 passed, 18 skipped, 0 failed.
The commercial repository carries the product's internal history: audits,
screenshots, briefs, decisions, pricing, publication drafts, personal data
and research runs. None of it belongs in the community edition.
- delete judging/, "UAT 3/", benchmarks/ (including LoCoMo data under
CC BY-NC), preflight-results/, gepa-phase-5/, notes/, decisions/,
.understand-anything/, .agents/, .lovable/, the legacy sidecar/ package
and waggle-cowork/ (an analysis of Claude Code sources)
- delete docs/ except docs/oss, the user guides (without the team-mode
guide), the reference, methodology, architecture, getting-started and
contributing pages, which P2.4 rewrites
- delete render.yaml, .parity-allowlist, the retired mind-parity and
sync-mind workflows and 29 one-off scripts nothing references
- delete tests that only checked removed files (sidecar package, Render
blueprint, UAT capture, benchmark harness isolation, internal audit and
signing docs); drop sidecar/ and benchmarks/ from the hive-mind subtree
split guard; prune ignore, attribute, lint and test globs
- replace the commercial README with an interim one for this edition
Source comments still point at some deleted docs; P2.2 part 2 cleans them
together with names and paths.
Self-learning is excluded from the community edition (ADR-0004); the
project owner reworks it separately. Cost tracking stays.
- agent: delete behaviour evolution (evolve-schema, iterative GEPA,
compose-evolution, evolution gates, orchestrator and deploy),
improvement and correction detection, AgentLearning, the eval-dataset
builder, optimization-capture and the agent_insights tool. Secret
detection moves from eval-dataset to secret-patterns. The LLM wiring
keeps only the Haiku adapter and judge call used by skill audit and
hygiene, so @ax-llm/ax is now declared by agent instead of server.
- server: delete /api/evolution/*, the evolution and optimizer services,
GEPA expansion of the first chat message, the prompt_optimization and
monthly_assessment jobs, the idle-session reviewer and its
session-reviewer persona, prompt injection of past corrections and
thumbs-down recorded as correction signals. Existing job rows of the
removed types are ignored, not deleted.
- core: delete OptimizationLogStore; stop re-exporting the improvement
signal and evolution run stores.
- web: delete the Evolution screen and the two job types in the
automation editor.
- delete packages/optimizer.
Also logs UF-027 (self-learning loops read data nothing collects) and
UF-028 (the Settings prompt-shape choice is ignored by chat).
KVARK is the commercial enterprise product and is not part of the community
edition. The agent's KVARK tools were never wired into the server, so
removing them changes no runtime behaviour beyond the routes listed here.
- agent: delete kvark-tools; CombinedRetrieval and formatCombinedResult now
merge workspace and personal memory only (no KVARK lane, no source
conflict note); connector substrate is always 'waggle'
- agent: promote_skill stops at workspace scope; team and enterprise scopes
answer "not available in this edition" instead of an upgrade prompt
- server: delete src/kvark and GET /api/marketplace/enterprise-packs
- marketplace: delete ENTERPRISE_PACKS
- compliance (EU AI Act dashboard, its templates and the KVARK template
seeded on every boot): delete core/src/compliance, the /api/compliance/*
routes, agent compliance-pdf and the Cockpit compliance panel. The
ai_interactions table lives in the memory substrate and goes with P1.5.
ADR-0007 maps today's memory usage onto the contract and mem0: profile
card and tasks move to waggle.db, learned facts stay in mem0, extraction
runs in the background in batches. Research notes cover what mem0 3.3.1
does and how 68 files use memory today. Two defects found on the way are
logged for the main project (UF-025, UF-026).
@waggle/memory-contract defines MemoryStore: remember (text, or a
conversation to extract facts from), search, list, count, get, update,
forget, forgetMatching and history over personal and workspace scopes,
with kinds, importance, source, origin and session in place of the old
content prefixes and named sessions. It ships an in-memory store and a
compatibility suite that every implementation runs.
@waggle/memory-mem0 implements the contract on mem0ai 3.3.1 in-process:
SQLite files in one directory, scope as mem0's agent_id, Waggle's own
embedder through mem0's langchain provider, fact extraction through any
OpenAI-compatible endpoint. mem0's telemetry and remote notice download
are switched off before the module loads; a test asserts that no request
is made. Memory survives a restart (tested).
New dependencies: mem0ai 3.3.1 (Apache-2.0, pinned exactly) with its
required peers compromise (MIT) and @cloudflare/workers-types (MIT OR
Apache-2.0); transitive openai (Apache-2.0), axios (MIT) and others under
MIT or BSD-2-Clause.
The root lint ignores apps/web, which has its own config and 49
pre-existing errors. Several e2e specs still exercise the removed tier and
Stripe endpoints; they are updated with the first e2e run.
Packages relied on npm hoisting for imports they never declared, so a
package installed on its own fails (UF-024).
- server: + @waggle/marketplace, @waggle/sdk, @waggle/weaver, zod (MIT),
@anthropic-ai/sdk (MIT), @ax-llm/ax (Apache-2.0);
- @waggle/hive-mind-shim-core and stripe (unused)
- agent: + pdf-parse (Apache-2.0); playwright-core (Apache-2.0) as an
optional peer, since it is loaded only when available
- worker: + @waggle/waggle-dance; web: + @waggle/shared
- launcher: + @waggle/server as a dev dependency (bundled by tsup);
- stripe; cli: - @waggle/weaver (unused); root: - stripe
scripts/check-workspace-deps.mjs (npm run check:deps) compares the
imports in each community-edition package's src/ with its manifest and
runs in CI. Two imports wait for P1.4 (optimizer, evolution wiring).
The community edition has no monetization (ADR-0001). Remove the tier
model (packages/shared/src/tiers.ts), the requireTier middleware and every
tier gate, GET/PATCH /api/tier and the trial start, and the Stripe
integration, which depends on the tier model.
Personal features a tier used to limit are open to everyone: unlimited
workspaces and connectors, every embedding provider including LiteLLM,
per-workspace costs, Approvals. The workspace session cap is one constant
(100). Team and enterprise surfaces are hidden instead of unlocked (team
dock zone, Settings Team/Enterprise/Plan tabs, workspace Team tab);
cloud sync reports unavailable, as it did for free users.
The sidecar ignores the tier keys in config.json but does not delete them:
both editions may still share ~/.waggle.
Web billing UI is deleted: plan cards, upgrade and trial modals, payment
success route, the KVARK sales block, tier 403 handling.
The previous contract described the commercial release: tiers, Stripe,
KVARK, the Windows Solo qualification gate and the hive-mind mirror sync.
It told agents to preserve monetization, which the community edition
removes. The new contract states the scope status of every package, the
conversion rules, the build and verification commands and the working
rules. CLAUDE.md now only adds Claude-specific notes.
ADR-0001..0006 capture the scope decisions: no monetization, memory
through a contract backed by local mem0 (knowledge graph deferred), team
mode postponed, cost tracking open and self-learning out, whatsapp-web.js
for WhatsApp, opt-in analytics. baseline.md records build, typecheck,
lint and unit test results on macOS before scope work starts.
The marketplace installer and the skill security cache resolve paths from
os.homedir() at import time, so npm run test wrote fixtures into the real
~/.waggle/security-cache and ~/.waggle/plugins/registry.json. Point HOME
and USERPROFILE at a temporary directory per worker and silence npm's
update notice that a fresh home triggers (UF-022). Log the four test
files that fail on macOS (UF-023).
The generic dist rule kept vendor/pptxgenjs/dist/pptxgen.cjs.js out of
git, so the sidecar failed to start on a fresh clone (UF-001). The file is
pptxgenjs 4.0.1 from npm, kept byte-identical via -text so its SHA-256
can be checked. The s*.png rule also hid the Settings visual baseline
(UF-021).
A non-streaming /api/pull holds the response headers until the download
ends, and Node fetch gives up on headers after five minutes, cancelling
every larger model pull. Request stream:true and read the NDJSON progress,
failing on an in-stream error or a missing success status (UF-008).
The Select dropdown is portaled to <body> with z-50, under the wizard's
z-[9999] scrim, so it was invisible yet clickable through the scrim.
Lift it to z-[10000] and cover the stacking order with a test (UF-007).