feat(summarize): report failures to Sentry + configurable retries

.dockerignore

@@ -0,0 +1,19 @@
+.git/
+node_modules/
+dist/
+*.tgz
+*.log
+.DS_Store
+.env
+.env.*
+!.env.example
+data/
+data-*/
+agentmemory-debug/
+.gstack/
+benchmark/
+eval/
+docs/
+READMEs/
+assets/
+CHANGELOG.md

.gitignore

@@ -19,10 +19,13 @@ data-*/
 agentmemory-debug/
 .gstack/
 
-# Lock files — never commit (see feedback_no_lockfiles memory)
+# Lock files — never commit (see feedback_no_lockfiles memory), except a
+# lockfile scoped ONLY to the Fly deploy build context, which is
+# deliberately committed for build reproducibility (deploy/fly/README.md).
 package-lock.json
 pnpm-lock.yaml
 yarn.lock
+!deploy/fly/package-lock.json
 integrations/hermes/__pycache__/
 
 # Eval reports (transient; published scorecards live in docs/benchmarks/)

deploy/fly/Dockerfile

@@ -2,9 +2,34 @@ ARG III_VERSION=0.11.2
 
 FROM iiidev/iii:${III_VERSION} AS iii-image
 
+# Builder stage: packs this branch's checked-out source into an npm
+# tarball at image-build time. The Dockerfile previously installed the
+# published `@agentmemory/agentmemory@${AGENTMEMORY_VERSION}` npm package,
+# which can't ship a fork patch until it's released upstream; an interim
+# local workaround (a manually-built, untracked tarball) covered that gap
+# but had no build-time check that it matched the checked-out source — a
+# fresh checkout with no local tarball would break, and a source change
+# with no tarball rebuild would silently ship stale code. Build context is
+# the repo root (see deploy/fly/README.md), so this stage sees the same
+# source tree `npm run build` would use locally.
+FROM node:22-slim AS builder
+WORKDIR /build
+# deploy/fly/package-lock.json is a lockfile scoped ONLY to this build
+# context, committed despite the repo-wide no-lockfile policy (see the
+# .gitignore exception) so this stage gets a reproducible dependency tree
+# instead of re-resolving semver ranges on every build. Regenerate it
+# deliberately (npm install --package-lock-only against the root
+# package.json) whenever a dependency is intentionally bumped.
+COPY package.json deploy/fly/package-lock.json ./
+RUN npm ci
+COPY tsconfig.json tsdown.config.ts ./
+COPY src/ ./src/
+COPY plugin/ ./plugin/
+COPY iii-config.yaml iii-config.docker.yaml docker-compose.yml .env.example LICENSE README.md AGENTS.md ./
+RUN npm run build && mkdir -p /build/out && npm pack --pack-destination /build/out
+
 FROM node:22-slim
 
-ARG AGENTMEMORY_VERSION=0.9.27
 ARG III_VERSION=0.11.2
 ARG III_SDK_VERSION=0.11.2
 
@@ -21,14 +46,26 @@ COPY --from=iii-image /app/iii /usr/local/bin/iii
 # is not refactored for yet). `npm install -g` ignores overrides, hence
 # the local prefix.
 WORKDIR /opt/agentmemory
+COPY --from=builder /build/out/*.tgz /opt/agentmemory/agentmemory.tgz
 RUN printf '{"name":"agentmemory-deploy","version":"1.0.0","private":true,"overrides":{"iii-sdk":"%s"}}\n' "${III_SDK_VERSION}" > package.json \
- && npm install "@agentmemory/agentmemory@${AGENTMEMORY_VERSION}" --omit=optional --no-fund --no-audit \
+ && npm install ./agentmemory.tgz --omit=optional --no-fund --no-audit \
  && ln -s /opt/agentmemory/node_modules/.bin/agentmemory /usr/local/bin/agentmemory
 
+# Optional NODE_OPTIONS preload for broad crash reporting: set
+#   NODE_OPTIONS="--require /opt/agentmemory/sentry-preload.cjs"
+# to capture uncaught exceptions/module-load failures that happen before
+# this package's own initSentry() (src/observability/sentry.ts) gets a
+# chance to run. Requires @sentry/node, already installed above as a
+# dependency of the packed tarball. No-op (just logs and exits the
+# require silently-successful) when SENTRY_DSN is unset. initSentry()
+# detects an already-initialized client from this preload and reuses it
+# instead of re-initializing over its tracesSampleRate/release.
+COPY deploy/fly/sentry-preload.cjs /opt/agentmemory/sentry-preload.cjs
+
 ENV AGENTMEMORY_III_VERSION=${III_VERSION} \
     TINI_SUBREAPER=1
 
-COPY --chmod=0755 entrypoint.sh /usr/local/bin/agentmemory-entrypoint.sh
+COPY --chmod=0755 deploy/fly/entrypoint.sh /usr/local/bin/agentmemory-entrypoint.sh
 
 EXPOSE 3111
 

deploy/fly/README.md

@@ -24,18 +24,25 @@ flow stays consistent:
 
 ```bash
 # 1. Install flyctl: https://fly.io/docs/flyctl/install/
-# 2. Pick your unique app name (and matching volume name):
+# 2. Pick your unique app name:
 export APP="agentmemory-$(whoami)"     # or any other globally-unique name
-export VOLUME="${APP//-/_}_data"       # Fly volume names can't contain '-'
-
-# 3. From this directory:
-fly launch --copy-config --no-deploy --name "$APP"
+export VOLUME="agentmemory_data"       # must exactly match [[mounts]].source
+                                        # in deploy/fly/fly.toml — NOT derived
+                                        # from $APP (that volume would never
+                                        # get attached to the app)
+
+# 3. From the REPO ROOT (not this directory) — the Dockerfile packs
+#    checked-out source into an npm tarball at build time, so the build
+#    context must be the whole repo, not deploy/fly/. (flyctl reads the
+#    Dockerfile path from fly.toml's own [build].dockerfile field, resolved
+#    relative to deploy/fly/ — there's no working CLI flag to override it.)
+fly launch --copy-config --no-deploy --name "$APP" --config deploy/fly/fly.toml
 
 # 4. Create the volume in the same region as the app:
 fly volumes create "$VOLUME" --region iad --size 1
 
-# 5. Deploy:
-fly deploy --app "$APP"
+# 5. Deploy (still from the repo root):
+fly deploy . --config deploy/fly/fly.toml --app "$APP"
 ```
 
 If `fly launch` reports the name is taken, pick another value for `$APP`,
@@ -115,6 +122,7 @@ inside the machine.
 fly ssh console --app "$APP"
 rm /data/.hmac
 exit
+fly machine list --app "$APP"        # copy the machine ID from the ID column
 fly machine restart <machine-id>
 fly logs --app "$APP" | grep AGENTMEMORY_SECRET=
 ```
@@ -132,6 +140,7 @@ To restore on a fresh machine:
 
 ```bash
 cat "$APP-YYYYMMDD.tar.gz" | fly ssh console --app "$APP" -C "tar xzf - -C /"
+fly machine list --app "$APP"        # copy the machine ID from the ID column
 fly machine restart <machine-id>
 ```
 
@@ -140,8 +149,8 @@ fly machine restart <machine-id>
 - Idle (machine stopped): the volume costs ~$0.15/GB/month. A 1 GB
   volume is roughly $0.15/month.
 - Active (machine running on `shared-cpu-1x` with 512 MB): about
-  $1.94/month if it ran 24/7; in practice `auto_stop_machines` keeps
-  that well under $1.
+  $3.69/month if it ran 24/7 (per Fly's current pricing page); in
+  practice `auto_stop_machines` keeps that well under $1.
 - Outbound bandwidth: 100 GB/month free on the Hobby plan, then $0.02/GB
   in North America / Europe.
 
@@ -152,15 +161,29 @@ See <https://fly.io/docs/about/pricing/> for the up-to-date rate card.
 - The volume lives in one region. To survive a region outage, create a
   second volume in another region and update `primary_region` after the
   failover, or take snapshots with `fly volumes snapshots create`.
-- The Dockerfile builds in the Fly Builder on every deploy — first
-  build is ~30 seconds; cached layers shrink rebuilds to under 10
-  seconds. Image is ~114 MB.
+- The Dockerfile builds in the Fly Builder on every deploy — the
+  `builder` stage compiles this branch's checked-out source and packs
+  it into an npm tarball, so the shipped image always matches the
+  commit you deploy from (no manually-built tarball to keep in sync).
+  First build is ~60 seconds (includes `npm ci` + `npm run build` +
+  `npm pack`); cached layers shrink rebuilds to well under that when
+  only `src/` changed. Image is ~124 MB.
 - First deploy lands on a **shared IPv4 + dedicated IPv6** by default
   (free). If you need a dedicated IPv4 for legacy clients without SNI,
   run `fly ips allocate-v4 --app "$APP"` — costs $2/month.
 - Cold-start (from machine launch to passing `/agentmemory/livez`) is
   ~9 seconds measured. `grace_period = "30s"` on the health check
-  gives a 3x safety margin.
-- Bump `AGENTMEMORY_VERSION` or `III_VERSION` in the Dockerfile to
-  upgrade. `fly deploy --build-arg AGENTMEMORY_VERSION=<x>` also works
-  for a one-off without editing the file.
+  gives more than 3x safety margin.
+- To upgrade `III_VERSION` (the `iii` engine binary), bump the
+  `ARG III_VERSION` default in `deploy/fly/Dockerfile` or pass
+  `--build-arg III_VERSION=<x>` on deploy. To upgrade the agentmemory
+  code itself, just commit the change and deploy — the builder stage
+  packs whatever is checked out, so there is no separate version arg to
+  bump.
+- `deploy/fly/package-lock.json` is a lockfile scoped only to this build
+  context (the repo otherwise gitignores lockfiles). It's committed for
+  build reproducibility and used by the builder stage's `npm ci`. When
+  you intentionally bump a dependency in `package.json`, regenerate it:
+  `npm install --package-lock-only && cp package-lock.json deploy/fly/package-lock.json && rm package-lock.json`
+  (run from the repo root). A stale lockfile makes `npm ci` fail loudly
+  at build time rather than silently drifting.

deploy/fly/fly.toml

@@ -12,6 +12,13 @@ app = "agentmemory"
 primary_region = "iad"
 
 [build]
+  # flyctl resolves this path relative to the DIRECTORY CONTAINING
+  # THIS CONFIG FILE (deploy/fly/), not the build context root —
+  # verified empirically 2026-07-13: "deploy/fly/Dockerfile" here
+  # doubled up to deploy/fly/deploy/fly/Dockerfile at build time.
+  # The build CONTEXT is still the repo root (see deploy/fly/README.md
+  # for the full `fly deploy .` invocation) — only this dockerfile
+  # path is config-relative.
   dockerfile = "Dockerfile"
 
 [[mounts]]