- install audit: build fixtures on WaggleDB, the store's real home; drop the
substrate SCHEMA_SQL parity half and the two MindDB-only recovery cases
for rows stranded in install_audit__mig_old.
- workspace manager: create() and ensure() no longer leave a workspace.mind;
getMindPath tests create the legacy file themselves. Drop the
MultiMindCache cases (reopen of a missing mind, post-resolution junction
swap): they tested the legacy cache, not the manager.
- transformers model load: use core's own in-process reranker.
The local sidecar has no /api/team/* routes since P1.1.
- Remove searchTeamMemory, getTeamMembers, getTeamActivity and
getTeamMessages from the adapter.
- Mission Control loses its Team and Activity tabs (both read the team
server) and keeps the fleet view, with its own load-error state.
- Drop the stale adapter mocks from the web tests; the Mission Control
test checks that no team tab is offered.
createTeamTools and createLocalTeamTools called /api/teams/* and
/api/team/* on a team server. Nothing registered them: not the
sidecar, the CLI, the team server or the worker.
packages/server/src/benchmarks held a LoCoMo failure-mode aggregator and
an LLM judge ensemble for the old memory engine benchmarks. Nothing in
the sidecar imports it and no npm script runs it; its inputs lived in
the root benchmarks/ directory that P2.2 already removed. Judge and eval
loops are also out of the community edition (AGENTS.md, rule 4).
The command bar could still offer "WaggleDance" and "Team" (governance)
as navigation targets, but those screens and app ids are gone from the
web app, so the frontend could not route them. Also drop a comment that
pointed at the removed WaggleDance routes.
- The WaggleDance routes left the sidecar with team mode, so the
x-waggle-run-token header no longer opens /api/waggle-dance/signal or
/api/waggle-dance/signals. A run token still authenticates only as a
Bearer on /v1/chat/completions (OpenClaw).
- Remove the session inactivity timeout. It switched on only in team
mode, when CLERK_SECRET_KEY was set.
Memory lives in the mem0 store and the knowledge graph is not in v1
(ADR-0002, ADR-0007).
- The CLI system prompt listed query_knowledge, a tool the community
edition does not register. It now lists search_all_workspaces, which
the orchestrator does provide.
- The default identity no longer claims "persistent memory (.mind file)"
or "knowledge graph queries".
Team mode is not part of the community edition (ADR-0003), so the CLI
no longer detects or switches to it.
- Remove the mode detector, the --local and --team flags, the Clerk
browser login (/login, /logout, /whoami) and /mode.
- Remove the /admin command and its team-server REST client.
- The welcome banner drops the Mode line.
- The installed-REPL test starts plain `waggle` and checks that the team
commands are gone.
secret-store.test.ts covers get, set, setAsync, writeBatch, delete,
deleteAsync, list and has through the interface. vault.test.ts keeps what is
specific to the file backend: encryption, the key file and the Windows ACL.
Connectors, the connector SDK and registry, setup-connectors, provider-env,
the LiteLLM runtime config and the sidecar's `vault` decoration only read and
write secrets, so they now take the SecretStore interface. VaultStore is still
constructed in the sidecar and the service startup, and its file format is
unchanged. AGENTS.md now points at the interface.
needsMigration and migrateToMultiMind moved a pre-M4 default.mind to
personal.mind at startup. The community edition has no installs with that
layout, so the module, its export, the startup step, its "migration" progress
phase and the default.mind first-run check go.
VaultStore.migrateFromConfig had two callers, both in the server. It is now
copyProviderKeysToVault in provider-env.ts, next to
migrateLegacyProviderKeysToVault, with the same rules: copy keys that have a
value, keep keys the vault already holds, store models and baseUrl as metadata.
Its tests moved from core to provider-env.test.ts.
getConnectorCredential, setConnectorCredential and setConnectorCredentialBundle
are now plain functions over a SecretStore in
packages/agent/src/connectors/connector-credentials.ts. The secret names,
metadata layout and the rule that tokens never go into plaintext metadata are
unchanged, so existing vault files keep working.
A bundle still lands as one update: SecretStore gains writeBatch, which
VaultStore applies with a single vault-file replacement.
Connector tests seed a Map-backed SecretStore instead of mocking VaultStore
methods; the helper tests from core and server moved to the agent package.
- tsconfig.build.json references the nine in-scope packages; each package
is composite and references the workspace packages it depends on.
build:packages is now `tsc -b tsconfig.build.json` instead of a chain
of cd commands.
- Every in-scope package exports its TypeScript entry under a `source`
condition next to the dist/ entries. Vitest, tsx (dev:server, the
Tauri dev launch, the e2e server), the sidecar and launcher bundles and
the web app resolve it, so none of them needs build:packages first or
can pick up a stale dist/. The vitest src aliases remain only for the
excluded packages that have no source export.
- Published packages with a files list (agent, cli, marketplace) pack
src/ too, so the source condition never points at a missing file.
- tsconfig.tests.json and `npm run typecheck:tests` type-check the test
suites of the in-scope packages. They report 426 existing errors, so
the CI step is advisory for now.
- The root test run leaves out the excluded and postponed packages
(hive-mind-*, memory-mcp, wiki-compiler, team-server, waggle-dance,
worker, weaver); they are not built either.
- CI runs the test job on Ubuntu, macOS and Windows with Node 22.23.2
(ADR-0008), and no longer builds the hook runtime and the memory MCP
server.
- Stop building and staging the hive-mind CLI, the coding-agent hook
packages and the memory MCP server: the hooks, Waggle Room and the
memory MCP server are hidden in v1 (ADR-0002) and all of them write to
the legacy .mind files.
- Drop sqlite-vec: nothing loads a SQLite extension since the legacy
memory engine left. bundle-native-deps no longer copies vec0, the
resource check no longer probes it, the Tauri shell no longer sets
WAGGLE_SQLITE_VEC_PATH and install.sh checks plain SQLite.
- Stage bundle roots that npm keeps inside a workspace. mem0ai is not
hoisted (it stays in packages/memory-mem0), so staging skipped it
silently and the packaged sidecar would fail on the first memory
access. Roots that are not installed are now listed in the output.
- js-yaml is a runtime dependency of Puppeteer's config loader since the
WhatsApp channel moved to whatsapp-web.js (ADR-0005); it leaves the
resource-check denylist. chromium-bidi stays unstaged: Puppeteer loads
it only for WebDriver BiDi sessions.
- @waggle/core no longer re-exports @waggle/hive-mind-core and no longer
depends on it.
- The server no longer opens personal.mind or a cache of workspace minds;
the legacy files are only read by the one-time copies of legacy data.
releaseWorkspace closes the workspace's sessions and detaches it from the
shared orchestrator.
- WorkspaceManager stops creating an empty workspace.mind.
- memory-center-legacy.ts had no callers left and is removed.
- build:packages no longer builds hive-mind-core and wiki-compiler; the
sidecar aliases the memory packages to source instead.
- mem0ai stays external in the sidecar and launcher bundles: it imports a
client for every provider it supports on demand, and bundling it failed on
the optional peers Waggle does not install. The launcher declares mem0ai
and no longer needs sqlite-vec.
- adapter: memory ids are strings; single-memory calls take only the id
(no ?mind= disambiguation). searchMemory takes the workspace as its own
parameter. New getMemoryCapabilities / setMemoryExtraction. Removed:
incrementFrameAccess, getMemoryOriginalSource, listSuppression,
allowReimport, getWeaverStatus, and the dead importPreview, importCommit
and getWikiWatermark. eraseMemory takes { id } or { sessionId }.
- memory-features reads GET /api/memory/capabilities for the knowledge
graph; history import, wiki, hooks, Room and the memory MCP stay hidden
(ADR-0002). useMemoryCapabilities re-renders the Memory screens.
- Settings > General has a Memory toggle for background fact extraction.
- Memory Center: the confidence filter, the erased-sources panel, the
original-source view and the Deprecated chip are gone; Erase forgets
every memory saved from the chat a memory came from and is offered only
for memories with a chat session. Memory kinds follow the contract.
- Trust views show short memory labels for the new string ids and drop the
per-row "not scored yet" badge (v1 stores no confidence).
- The Weaver maintenance view and the Cockpit weaver card are removed
(the weaver package and its routes are deleted).
- Timeline uses the contract kinds and maps important/temporary
importance.
Logs UF-033 (search scope sent as the workspace id), UF-034 (importance
mapping) and UF-035 (restore over open databases, open).
The memory HTTP API reads and writes server.memory instead of the legacy
MultiMind, FrameStore, SessionStore, KnowledgeGraph, MindErasure,
SuppressionStore and RawArchive objects.
- memory.ts: search (personal, workspace, all, global), frames list, write,
edit and delete, stats and quick capture on the contract. Ids are strings.
The access-count route is gone. New GET /api/memory/capabilities and
PUT /api/memory/settings { extraction } for background fact extraction.
- memory-center.ts: list, get, create, patch, archive, confirm, trace,
delete, erase and merge on the contract. Archive sets importance
'temporary' (skipped by recall) and remembers the old importance; confirm
stamps metadata.confirmedAt; 'unreviewed' means extracted in the
background and not confirmed. Erase forgets one memory or every memory of
a chat session. The original-source and suppression routes are removed:
there is no raw archive and no harvest to re-import from.
- knowledge.ts returns an empty graph (not part of v1).
- export.ts writes memories/personal.json and memories/workspace-<id>.json
from the contract.
- identity.ts loses a dead resolver; the erase safety check recognizes
waggle.db.
- memory-records.ts holds the scope, status and creation-time helpers.
- normalizeToMemory stays in memory-center-legacy.ts until home.ts and
artifacts.ts read memory through the contract.
- @waggle/shared MemoryKind follows the contract's kinds; Memory gains
origin and sessionId.
Tests: new memory-routes and memory-center tests on the in-memory store;
local-mode and data-export tests seed memory through the contract; tests of
removed behaviour (source, suppression, sticky erasure, graph projection,
legacy normalizer, old erase and stats) are deleted.
The comprehensive and hard-persistence suites open a data directory the
way the REPL does: mem0 in <dir>/memory (word embedder, no model download)
and waggle.db for the profile card and tasks. Persistence across sessions
now also checks identity and tasks after reopening.
The packaged-runtime tests install @waggle/memory-contract and
@waggle/memory-mem0 with the CLI (the weaver package is gone) and check
that the REPL creates the memory store and waggle.db instead of
default.mind.
Deleted: T4 and T18 (knowledge graph entities and their persistence) and
the knowledge-graph steps of the hard-persistence run; the graph is not in
v1 (ADR-0002).
Rewrites the orchestrator, recall, compaction, save_memory and assembler
suites for MemoryStore (createInMemoryMemoryStore; mem0 with the word
embedder for the M3b integration test), string ids, async prompt building
and setWorkspace(). Creation times are set with fake timers, since a write
cannot set its own. Stores of Waggle's own data (identity, traces) are built
on WaggleDB; the session-manager stubs use the new create() signature.
Kept behaviour: importance lane (cap, active scope, temporary exclusion),
profile and event lanes, date window and its fallback, catch-up, injection
blocking of recall and preloaded context, autosave safety, decision consent,
compaction update-in-place, reranker wiring.
New tests: recall with a workspace active versus personal only;
fetchRecentMemories ordering; toRecallFrame; context loader; save_memory
routing, dedup, rate limit, dramatic-claim downgrade and contradiction
scope; search_memory and search_all_workspaces; pattern write-back kinds,
scopes and sign gate; read_other_workspace; reranker off unless
WAGGLE_RERANKER=1, wider rerank pool, failure fallbacks.
Deleted, because the behaviour is gone:
- w46-rawdetail-recall.test.ts: the raw-turn (RAWDETAIL) recall lane.
- orchestrator.test.ts, query_knowledge (3 tests) and the recall-hardening
M4 "Key Knowledge" test: the knowledge graph is not in v1 (ADR-0002).
- orchestrator-context-frames "profile override": scoring profiles.
- Assertions on I/P frame types, GOP sessions, FTS rows, the "Memory Facts"
lane, cognify stages and session/entity counts were replaced by the
equivalent contract-level checks.
The pattern write-back pushed a style note into its "saved" list even when
the memory store answered 'unchanged', so the chat route announced
"Auto-saved 1 memory" on every repeat of the same style signal. The legacy
code checked for the note before writing; the contract port lost that.
A style note that already exists is now neither written nor reported.
Other write-back saves keep reporting repeats, as before the port.
Not logged in upstream-fixes.md: the defect came with the P1.5 port and
does not exist in the commercial product.
- Workspace sessions, fleet runs, agent groups, loops, the file indexer,
external tool runs, chat and the health endpoint are checked against the
memory store instead of .mind files. Assertions on mind-cache pins and
releases go with the pins; the drain and cancellation checks stay.
- Delete the sticky-erasure test (harvest import was removed) and the
session-distillation test (distillation on workspace activation was
removed).
- The saved-history retry test checks the conversation only: the failed
user turn is kept in workspace memory by design (#3).
Chat collaboration writes run results to memory in the background and
records the memory refs when the writes finish. That update recomputed the
parent Room, so a workflow Room turned 'completed' while its synthesis step
was still running and could no longer be cancelled. A memory-refs update
never changes a run's status, so it no longer recomputes the parent.
The collaboration tests use an in-memory memory store and wait for the
background writes before reading memory refs.
- Artifact Center related search lists personal and workspace memories
through server.memory and finds related chat sessions in the session
files (UF-034).
- /api/agent/run searches the requested workspace's memory, or personal
memory when no known workspace is given.
- Executor briefs search only the workspace's own memory and leave out
temporary memories in the query; route proposals no longer borrow a
mind from the cache.
- Onboarding treats any personal memory as evidence of prior use.
- The service no longer creates personal.mind on a fresh install, and
isFirstRun also looks for waggle.db.
- Workspace state, the "Workspace Now" block, the home briefing and the
proactive handlers read the workspace's memory through server.memory:
decisions are memories of kind 'decision', critical ones or ones worded
as decisions; summaries list memories most important first. They no
longer open workspace .mind files, so the builders are async and
ProactiveContext carries `memory` instead of `getWorkspaceMindDb`.
- The home briefing counts extracted personal facts that are neither
confirmed nor archived as awaiting review (memory-status.ts).
- Workspace routes count memories through the contract, seed template
starter memories with remember() in the background (origin
'workspace-template'), export memories through list(), and deleting a
workspace forgets its memory scope and clears its awareness items before
the directory is removed; the delete is refused if forgetting fails.
agentState.releaseWorkspace() replaces closeWorkspaceMind for callers.
- The command palette searches personal memory and the active workspace's
memory (UF-033).
- chat.ts drops the MindDB-flake diagnostics; the mind cache they guarded
no longer serves chat turns.
mem0's delete kept the deleted text in memory_history, and fact extraction
keeps the last messages of each scope in the same file. forget now purges
the memory's history rows, and forgetting a whole scope also drops the
conversation messages mem0 kept for it.
signal-emitter-integration.test.ts posted hook signals to /api/waggle-dance/signal on a real sidecar. The route left the sidecar with team mode; the emitter's own behaviour, fail-open included, stays covered in hive-mind-shim-core.
The previous commit dropped pg together with the team-only externals, but pg belongs to the agent's Postgres connector, which loads it on demand. Keep it external as before.
Every sidecar start outside the desktop app looked for Python, probed
each interpreter for litellm and, when it found one, started a LiteLLM
child. The community edition must not need Python, so LiteLLM is now
opt-in: startService starts it only when config.json contains
"litellm": { "enabled": true } (WaggleConfig.isLiteLLMEnabled). Otherwise
there is no Python lookup, no LiteLLM health probe and no child process,
and the built-in provider proxy serves every request, as it already did
in the desktop app. --skip-litellm and WAGGLE_SKIP_LITELLM=1 still force
it off.
The lookup for a bundled resources/python/python.exe is removed: no
build ships one. When LiteLLM is enabled it runs on the user's own
Python from PATH.
Team mode stays in the repository but is not built or shipped (ADR-0003).
- The team server (index.ts, config, db, daemons, plugins, proactive,
routes, scheduler, services, ws, drizzle migrations) and its tests move
from @waggle/server to packages/team-server (@waggle/team-server,
private). It keeps its own copies of the logger and the agent-group
executor, so it does not import @waggle/server. @clerk/fastify, bullmq,
ioredis, postgres, drizzle-orm, drizzle-kit, cron-parser and
@waggle/waggle-dance leave @waggle/server; the launcher drops the same
team-only dependencies.
- build:packages no longer builds waggle-dance or the worker: nothing in
the community edition imports them, and the worker's project reference
would otherwise build the team server. The worker now imports from
team-server; the team Dockerfile builds both explicitly.
- The sidecar no longer starts the team server when DATABASE_URL is set,
and loses the team routes (/api/team/*, /api/teams), the WaggleDance
routes and bridge (/api/waggle-dance/*), the Team server binding and
egress guard, the WebSocket team client, the TeamSync cache and pushes,
team audit forwarding, team workspace registration and the team
capability-policy lookup in chat. The in-process room signal bus stays:
agent groups, fleet runs and external tool runs use it.
middleware/audit.ts was dead and is deleted.
- @waggle/core loses TeamSync and the teamServer part of WaggleConfig. A
teamServer key in config.json is ignored and kept on save, as P0.1 does
for the tier keys. Workspaces lose teamServerUrl and teamUserId; teamId
and teamRole stay because workspaces from a shared data directory still
carry them and the viewer write guards read them. POST /api/workspaces
ignores team fields.
- The sidecar bundle no longer marks the team packages as external.
The agent still references the TeamSync type in orchestrator.ts and
pattern-write-back.ts; those files are being rewritten in P1.5 and drop
it there.
Logs UF-030 and UF-031 (team sync and team policy defects found while
removing the code).
Nothing created an S3FileStore outside team workspaces; team storage is
postponed with team mode (ADR-0003), so such workspaces use virtual storage.
@aws-sdk/client-s3 leaves core's dependencies.
Every VaultStore construction started PowerShell to re-apply the key ACL.
The ACL does not change while the process runs; a created key is still
written and hardened by the same PowerShell call.
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.