diff --git a/.changeset/yellow-windows-swim.md b/.changeset/yellow-windows-swim.md
new file mode 100644
index 0000000000..6f1f20ec31
--- /dev/null
+++ b/.changeset/yellow-windows-swim.md
@@ -0,0 +1,6 @@
+---
+---
+
+Add a shared image for the DAO API databases that derives `max_connections` and
+`effective_cache_size` from the container's cgroup memory limit at boot.
+Infra-only — no workspace package changes.
diff --git a/infra/dao-api-db/Dockerfile.dao-api-db b/infra/dao-api-db/Dockerfile.dao-api-db
new file mode 100644
index 0000000000..2483a0814f
--- /dev/null
+++ b/infra/dao-api-db/Dockerfile.dao-api-db
@@ -0,0 +1,23 @@
+# The fleet spans three Postgres majors (16 x8, 17 x4, 18 x5), so the version
+# has to come from the service, not the file. Railway exposes service variables
+# as build args, so each DB service sets PG_MAJOR to match the major that
+# initialised its volume.
+#
+# Deliberately no default: an unset PG_MAJOR fails the *build*, which leaves the
+# currently-running container serving. A wrong default would instead build fine
+# and then crashloop on "database files are incompatible with server" — a louder
+# failure at a much worse moment.
+ARG PG_MAJOR
+FROM ghcr.io/railwayapp-templates/postgres-ssl:${PG_MAJOR}
+
+# The base image's ENTRYPOINT (wrapper.sh) sets up SSL certs and pgbackrest, so
+# we don't replace it — entrypoint.dao-api-db.sh derives a couple of settings from
+# the container's memory limit and then execs wrapper.sh with them appended.
+COPY entrypoint.dao-api-db.sh /entrypoint.sh
+RUN chmod +x /entrypoint.sh
+ENTRYPOINT ["/entrypoint.sh"]
+
+# Declaring ENTRYPOINT in a derived image resets the base image's CMD to empty,
+# so it has to be repeated verbatim. This is what `ps` shows in a running
+# stock DB container: `wrapper.sh postgres --port=5432`.
+CMD ["postgres", "--port=5432"]
diff --git a/infra/dao-api-db/dao-api-db.railway.toml b/infra/dao-api-db/dao-api-db.railway.toml
new file mode 100644
index 0000000000..78182010bd
--- /dev/null
+++ b/infra/dao-api-db/dao-api-db.railway.toml
@@ -0,0 +1,10 @@
+[build]
+builder = "DOCKERFILE"
+dockerfilePath = "infra/dao-api-db/Dockerfile.dao-api-db"
+
+[deploy]
+# No healthcheckPath: Postgres speaks the wire protocol, not HTTP, so Railway's
+# HTTP probe would never pass. Liveness is covered by postgres-exporter's pg_up,
+# which feeds the PostgreSQLUnavailable alert in infra/monitoring/alerts.yml.
+restartPolicyType = "ON_FAILURE"
+restartPolicyMaxRetries = 10
diff --git a/infra/dao-api-db/entrypoint.dao-api-db.sh b/infra/dao-api-db/entrypoint.dao-api-db.sh
new file mode 100644
index 0000000000..26543d767f
--- /dev/null
+++ b/infra/dao-api-db/entrypoint.dao-api-db.sh
@@ -0,0 +1,93 @@
+#!/bin/bash
+# Derive the memory-dependent Postgres settings from the container's own cgroup
+# limit, then hand off to the base image's entrypoint with them appended as -c
+# flags.
+#
+# Why derive instead of hardcode: the DAO databases run on four different memory
+# tiers (2 / 4 / 6 / 8 GB) off this one image, and Railway plan changes move a
+# service between tiers without touching the repo. Reading the limit at boot
+# keeps the settings correct through both.
+#
+# Why -c flags instead of ALTER SYSTEM: command-line options outrank
+# postgresql.auto.conf, so this also neutralises the hand-applied
+# `max_connections = 500` currently baked into every DAO volume — without
+# writing to the volumes, and revertible by pointing a service back at the
+# stock image.
+set -euo pipefail
+
+# Smallest tier in the fleet. Used when the limit is unreadable or unset ("max"),
+# because under-provisioning degrades and over-provisioning OOMs.
+readonly FALLBACK_BYTES=2000000000
+
+# $1 = memory limit in bytes. Echoes "<max_connections> <effective_cache_size_mb>".
+derive() {
+  local bytes=$1 gb mc ecs_mb
+
+  # Non-numeric covers both "max" (no limit set) and a missing cgroup file.
+  if [[ ! $bytes =~ ^[0-9]+$ ]] || (( bytes == 0 )); then
+    bytes=$FALLBACK_BYTES
+  fi
+
+  # Round to the nearest GB: cgroup reports 1999998976 for a 2 GB tier, and
+  # truncating that to 1 would put every tier a notch too low.
+  gb=$(( (bytes + 500000000) / 1000000000 ))
+
+  # ~20 backends per GB, floored so Ponder's two default-30 pools always fit and
+  # capped so a large tier can't authorise more unreclaimable memory than the
+  # page cache can spare. 500 backends against the 2 GB tier is ~2.5 GB of
+  # non-reclaimable memory — the OOM path this exists to close.
+  mc=$(( gb * 20 ))
+  (( mc < 75 )) && mc=75
+  (( mc > 200 )) && mc=200
+
+  # Planner hint only — allocates nothing. shared_buffers stays at its 128 MB
+  # default, so effectively all of this is kernel page cache.
+  ecs_mb=$(( bytes * 65 / 100 / 1048576 ))
+
+  echo "$mc $ecs_mb"
+}
+
+# Asserts the tiers actually in the fleet plus both clamp boundaries.
+# Run locally or in CI with: bash entrypoint.dao-api-db.sh --self-test
+self_test() {
+  local failed=0
+  check() {
+    local got
+    got=$(derive "$1")
+    if [[ $got != "$2" ]]; then
+      echo "FAIL: derive($1) = '$got', expected '$2'" >&2
+      failed=1
+    fi
+  }
+
+  check 1999998976 "75 1239"    # 2 GB tier (nouns, shutter) — floored
+  check 3999997952 "80 2479"    # 4 GB tier (ens, compound, uniswap@dev)
+  check 5999996928 "120 3719"   # 6 GB tier (aave)
+  check 8000000000 "160 4959"   # 8 GB tier (uniswap@prod)
+  check 1000000000 "75 619"     # below the floor
+  check 16000000000 "200 9918"  # above the cap
+  check "max" "75 1239"         # no limit set -> fallback
+  check 0 "75 1239"             # unreadable -> fallback
+
+  (( failed )) && { echo "self-test FAILED" >&2; return 1; }
+  echo "self-test OK"
+}
+
+if [[ ${1:-} == "--self-test" ]]; then
+  self_test
+  exit $?
+fi
+
+limit_bytes=$(cat /sys/fs/cgroup/memory.max 2>/dev/null || echo max)
+read -r max_connections effective_cache_size_mb <<<"$(derive "$limit_bytes")"
+
+# Escape hatches, in case one database needs to deviate without a redeploy of
+# the image.
+max_connections=${PG_MAX_CONNECTIONS:-$max_connections}
+effective_cache_size=${PG_EFFECTIVE_CACHE_SIZE:-${effective_cache_size_mb}MB}
+
+echo "[anticapture] memory.max=${limit_bytes} -> max_connections=${max_connections} effective_cache_size=${effective_cache_size}"
+
+exec /usr/local/bin/wrapper.sh "$@" \
+  -c max_connections="$max_connections" \
+  -c effective_cache_size="$effective_cache_size"