factory-ai 1.3.0 → 1.5.0

package/CHANGELOG.md

@@ -4,12 +4,32 @@ All notable changes follow semantic versioning and the Keep a Changelog structur
 
 ## [Unreleased]
 
+## [1.5.0] - 2026-07-13
+
+### Added
+
+- Cross-platform Bubble Tea/Lip Gloss operator client with direct Azure Blob snapshots, verified release binaries, scrolling, and safe Node action fallback.
+- Automatic Azure and Bedrock context compaction with immutable objective retention and recent tool evidence.
+- Hierarchical `AGENTS.md` discovery, preconfigured project instructions, and bounded `.agent-factory` context.
+- Durable worktree crash recovery, parallel read-only tool batches, live activity timelines, retry/error telemetry, and stale-agent watchdog recovery.
+- Post-setup `factory models show|set|reset` and `factory configure models` workflows with validated routes, atomic runtime rollback, and update-safe persistence.
+- Open-source harness adoption research and executable Go CI gates.
+- Token-budgeted repository maps, safe OpenTelemetry GenAI records, Promptfoo/Inspect/Harbor evaluation contracts, typed lifecycle hooks, durable approval interrupts, optional ACP intake, and signed extension verification.
+
+### Security
+
+- Read-only role filesystems and command policy, deterministic pre-push Gitleaks enforcement, typed scanner release blockers, terminal-safe Go rendering, and structured allowlisted operator logs.
+- Repository-scoped MCP memory; legacy unscoped memory is retained as `legacy-unscoped-knowledge-graph.jsonl` instead of being mixed into new repositories.
+- Model credentials now enter agent containers through a consumed stdin envelope rather than environment variables; a persistent Docker firewall blocks cloud metadata access.
+- Release publishing now binds tags to package versions, requires full gates and npm provenance, and attests immutable Go binaries.
+
+## [1.4.0] - 2026-07-13
+
 ### Added
 
-- Full-screen terminal administration console.
-- Telegram remote objective intake with chat allowlisting.
-- Azure and Bedrock provider wizard.
-- Durable evaluation, analytics, scaling, policy, extension, and recovery roadmap.
+- Free local semantic code retrieval with pinned Ollama embeddings and retained Qdrant vectors.
+- Commit-aware indexing, bounded chunking, top-k context injection, and failure fallback.
+- Modern agent-harness capability matrix and explicit completion criteria.
 
 ## [1.3.0] - 2026-07-13
 

package/Dockerfile.worker

@@ -1,4 +1,4 @@
-FROM node:20.20.2-bookworm-slim
+FROM node:22.17.0-bookworm-slim
 
 RUN apt-get update \
   && apt-get install -y --no-install-recommends git ripgrep ca-certificates \

package/README.md

@@ -73,7 +73,13 @@ factory ui
 | `factory issue OWNER/REPO NUMBER` | Turn a GitHub issue into a tracked objective |
 | `factory telegram configure` | Configure allowlisted Telegram remote intake |
 | `factory dashboard` | Objectives, agents, models, queue, DLQ, and Azure cost |
-| `factory init PATH` | Create durable repo-local project context |
+| `factory init PATH` | Create `AGENTS.md` and durable repo-local project context without overwriting existing files |
+| `factory models show` | Show effective role-to-model routes |
+| `factory models set ROLE PROVIDER/MODEL` | Validate and atomically switch a role to a new model |
+| `factory models reset ROLE` | Restore a role to the versioned default |
+| `factory configure models` | Interactively change model routing after setup |
+| `factory acp REQUEST.json` | Submit a strictly validated optional ACP objective |
+| `factory extension verify MANIFEST ARTIFACT PUBLIC_KEY` | Verify a signed extension manifest and immutable artifact digest |
 | `factory doctor` | Services, storage, memory, and host health |
 | `factory queue` | Queue and dead-letter depth |
 | `factory logs` | Consolidated service logs |
@@ -131,6 +137,7 @@ Factory AI minimizes tokens before relying on cheaper models:
 - File reads are line-ranged and bounded; listings, commands, MCP output, memory, and scanner evidence are truncated with continuation hints.
 - Read-only roles do not receive write-tool schemas.
 - Planner memory is compact, repository-scoped, and limited to recent verified events.
+- Ollama generates local embeddings and Qdrant retrieves only top-ranked code chunks, avoiding embedding API charges and whole-repository prompts.
 - Dashboard and TUI track input, cached-input, and output tokens by model.
 - The planner is instructed to produce the smallest valid DAG, avoiding duplicate agents.
 
@@ -156,7 +163,7 @@ Every repository gets two memory layers:
 
 Built-in progressive skills include `/goal`, `/loop`, project context, systematic debugging, TDD, verification, security review, dependency security, browser verification, release discipline, and token efficiency. Context7 and Playwright MCPs are pinned and role-scoped.
 
-Use `factory init PATH` to create `.agent-factory/` project, architecture, commands, decisions, risks, and handoff files without overwriting existing context.
+Use `factory init PATH` to create a safe root `AGENTS.md` plus `.agent-factory/` project, architecture, commands, decisions, risks, and handoff files without overwriting existing context. The runtime discovers repository `AGENTS.md` instructions for planners and workers. Active Azure and Bedrock conversations compact automatically after their configured token threshold while preserving bounded recent tool evidence.
 
 ## Credentials
 
@@ -255,6 +262,7 @@ npm pack --dry-run
 | [CONTRIBUTING.md](CONTRIBUTING.md) | Development and verification contract |
 | [HANDOFF.md](HANDOFF.md) | Team/friend transfer context |
 | [docs/COMPARISON.md](docs/COMPARISON.md) | Honest comparison with paid alternatives |
+| [docs/HARNESS_PARITY.md](docs/HARNESS_PARITY.md) | Feature parity across modern agent harnesses |
 | [ROADMAP.md](ROADMAP.md) | Planned platform and ecosystem work |
 | [GOVERNANCE.md](GOVERNANCE.md) | Decision and release governance |
 | [SUPPORT.md](SUPPORT.md) | Community support process |

package/ROADMAP.md

@@ -4,7 +4,6 @@
 
 - Evaluation Lab with model promotion and rollback evidence.
 - Enhanced TUI timeline, diffs, files, checks, and safe takeover leases.
-- Telegram remote objective status and completion notifications.
 - Multi-repository objective schema and release groups.
 
 ## Platform

package/bin/factory

@@ -10,6 +10,8 @@ VM=${FACTORY_VM:-agent-factory-vm}
 NAMESPACE=${FACTORY_SERVICE_BUS:-}
 VAULT=${FACTORY_KEY_VAULT:-}
 STORAGE=${FACTORY_STORAGE_ACCOUNT:-}
+FACTORY_NAME=${FACTORY_NAME:-Factory AI}
+FACTORY_PURPOSE=${FACTORY_PURPOSE:-Ship secure reviewed software continuously}
 
 remote() {
   az vm run-command invoke --resource-group "$RESOURCE_GROUP" --name "$VM" \
@@ -21,6 +23,9 @@ usage() {
     'Factory AI operator console' \
     'Usage: factory <command>' \
     '  setup' \
+    '  configure models' \
+    '  models show | set ROLE PROVIDER/MODEL | reset ROLE' \
+    '  acp REQUEST.json | extension verify MANIFEST ARTIFACT PUBLIC_KEY' \
     '  github status | connect ORG | transfer OWNER/REPO ORG' \
     '  telegram status | configure | start | stop' \
     '  submit <owner/repo> <objective>' \
@@ -32,9 +37,99 @@ usage() {
     '  pause | resume'
 }
 
-command=${1:-help}
+command=${1:-ui}
 shift || true
+
+apply_runtime_model() {
+  local variable=$1 value=${2:-} mode=${3:-set} script settings
+  script="set -euo pipefail; exec 9>/run/lock/factory-ai-config.lock; flock -x 9; first=/etc/agent-factory.env; second=/etc/agent-factory-control.env; backup_first=\$(mktemp); backup_second=\$(mktemp); cp --preserve=mode,ownership \"\$first\" \"\$backup_first\"; cp --preserve=mode,ownership \"\$second\" \"\$backup_second\"; rollback() { install -m 0600 -o root -g root \"\$backup_first\" \"\$first\"; install -m 0600 -o root -g root \"\$backup_second\" \"\$second\"; systemctl restart agent-factory-control agent-factory-worker agent-factory-release || true; rm -f \"\$backup_first\" \"\$backup_second\"; }; trap rollback ERR TERM INT; for file in \"\$first\" \"\$second\"; do tmp=\$(mktemp); grep -v '^${variable}=' \"\$file\" > \"\$tmp\" || true; if [[ '$mode' == set ]]; then printf '${variable}=%s\\n' '$value' >> \"\$tmp\"; fi; install -m 0600 -o root -g root \"\$tmp\" \"\$file\"; rm -f \"\$tmp\"; done; systemctl restart agent-factory-control agent-factory-worker agent-factory-release; systemctl is-active --quiet agent-factory-control agent-factory-worker agent-factory-release; trap - ERR TERM INT; rm -f \"\$backup_first\" \"\$backup_second\""
+  printf -v script 'bash -c %q' "$script"
+  settings=$(jq -n --arg command "$script" '{commandToExecute:$command}')
+  az vm extension set --resource-group "$RESOURCE_GROUP" --vm-name "$VM" --name FactoryAIConfigure \
+    --publisher Microsoft.Azure.Extensions --version 2.1 --force-update --protected-settings "$settings" --output none
+}
+
+update_local_model() {
+  local variable=$1 value=${2:-} mode=${3:-set} temporary
+  mkdir -p "$(dirname "$CONFIG_FILE")"
+  temporary=$(mktemp "${CONFIG_FILE}.XXXXXX")
+  if [[ -f $CONFIG_FILE ]]; then grep -v "^${variable}=" "$CONFIG_FILE" > "$temporary" || true
+  elif [[ -f $LEGACY_CONFIG_FILE ]]; then grep -v "^${variable}=" "$LEGACY_CONFIG_FILE" > "$temporary" || true; fi
+  if [[ $mode == set ]]; then printf '%s=%q\n' "$variable" "$value" >> "$temporary"; fi
+  chmod 0600 "$temporary"
+  mv "$temporary" "$CONFIG_FILE"
+}
+
+configure_model_transaction() {
+  local variable=$1 value=${2:-} mode=${3:-set}
+  MODEL_CONFIG_BACKUP=$(mktemp)
+  MODEL_CONFIG_EXISTED=false
+  if [[ -f $CONFIG_FILE ]]; then cp "$CONFIG_FILE" "$MODEL_CONFIG_BACKUP"; MODEL_CONFIG_EXISTED=true; fi
+  rollback_local_model() {
+    if [[ $MODEL_CONFIG_EXISTED == true ]]; then mv "$MODEL_CONFIG_BACKUP" "$CONFIG_FILE"; else rm -f "$CONFIG_FILE" "$MODEL_CONFIG_BACKUP"; fi
+  }
+  trap rollback_local_model EXIT INT TERM
+  update_local_model "$variable" "$value" "$mode"
+  apply_runtime_model "$variable" "$value" "$mode"
+  rm -f "$MODEL_CONFIG_BACKUP"
+  trap - EXIT INT TERM
+}
+
 case "$command" in
+  acp)
+    request=${1:?ACP request JSON file required}
+    SERVICE_BUS_NAMESPACE="$NAMESPACE" CONTROL_QUEUE=control-events FACTORY_ACP_ENABLED=true exec node "$FACTORY_ROOT/src/acp-cli.js" "$request"
+    ;;
+  extension)
+    [[ ${1:-} == verify ]] || { printf 'Usage: factory extension verify MANIFEST ARTIFACT PUBLIC_KEY\n' >&2; exit 2; }
+    exec node "$FACTORY_ROOT/src/extension-cli.js" "${2:?manifest required}" "${3:?artifact required}" "${4:?public key required}"
+    ;;
+  configure)
+    area=${1:-models}
+    [[ $area == models ]] || { printf 'Supported configuration area: models\n' >&2; exit 2; }
+    "$0" models show
+    read -r -p 'Role to change (or Enter to leave unchanged): ' role
+    [[ -n $role ]] || exit 0
+    read -r -p 'New provider/model route (or "default"): ' route
+    if [[ $route == default ]]; then exec "$0" models reset "$role"; fi
+    exec "$0" models set "$role" "$route"
+    ;;
+  models)
+    action=${1:-show}
+    role=${2:-}
+    route=${3:-}
+    case "$action" in
+      show)
+        for current in scout planner builder tester debugger reviewer security release; do
+          upper=$(printf '%s' "$current" | tr '[:lower:]' '[:upper:]')
+          variable="FACTORY_MODEL_${upper}"
+          case "$current" in
+            scout) default=azureai-textved/factory-gpt-5-4-nano ;;
+            tester) default=azureai-responses/gpt-5.4 ;;
+            builder) default=azureai-textved/gpt-5.5 ;;
+            *) default=azureai-textved/gpt-5.6-sol ;;
+          esac
+          printf '%-10s %s%s\n' "$current" "${!variable:-$default}" "$([[ -n ${!variable:-} ]] && printf ' (override)' || true)"
+        done
+        ;;
+      set)
+        [[ $role =~ ^(scout|planner|builder|tester|debugger|reviewer|security|release)$ ]] || { printf 'Invalid model role\n' >&2; exit 2; }
+        node --input-type=module -e "import { validateModelRoute } from 'file://$FACTORY_ROOT/src/routing.js'; validateModelRoute(process.argv[1])" "$route"
+        upper=$(printf '%s' "$role" | tr '[:lower:]' '[:upper:]')
+        variable="FACTORY_MODEL_${upper}"
+        configure_model_transaction "$variable" "$route" set
+        printf '%s now routes to %s. Runtime services are healthy.\n' "$role" "$route"
+        ;;
+      reset)
+        [[ $role =~ ^(scout|planner|builder|tester|debugger|reviewer|security|release)$ ]] || { printf 'Invalid model role\n' >&2; exit 2; }
+        upper=$(printf '%s' "$role" | tr '[:lower:]' '[:upper:]')
+        variable="FACTORY_MODEL_${upper}"
+        configure_model_transaction "$variable" "" reset
+        printf '%s restored to the versioned default. Runtime services are healthy.\n' "$role"
+        ;;
+      *) printf 'Usage: factory models show | set ROLE PROVIDER/MODEL | reset ROLE\n' >&2; exit 2 ;;
+    esac
+    ;;
   telegram)
     action=${1:-status}
     case "$action" in
@@ -74,7 +169,34 @@ case "$command" in
     exec "$0" submit "$repo" "$objective"
     ;;
   ui)
-    FACTORY_RESOURCE_GROUP="$RESOURCE_GROUP" FACTORY_VM="$VM" FACTORY_SERVICE_BUS="$NAMESPACE" FACTORY_KEY_VAULT="$VAULT" FACTORY_STORAGE_ACCOUNT="$STORAGE" node "$FACTORY_ROOT/src/tui.js"
+    if [[ ! -f $CONFIG_FILE && ! -f $LEGACY_CONFIG_FILE ]]; then
+      "$0" setup
+      exec "$0" ui
+    fi
+    version=$(node -p "require('$FACTORY_ROOT/package.json').version")
+    os=$(uname -s | tr '[:upper:]' '[:lower:]')
+    arch=$(uname -m)
+    [[ $arch == x86_64 ]] && arch=amd64
+    [[ $arch == aarch64 ]] && arch=arm64
+    cache="$HOME/.cache/factory-ai/$version"
+    binary="$cache/factory-ui-$os-$arch"
+    if [[ ! -x $binary ]]; then
+      mkdir -p "$cache"
+      base="https://github.com/itsvedantkumar/factory-ai/releases/download/v$version"
+      if curl -fsSL "$base/factory-ui-$os-$arch" -o "$binary.tmp" && curl -fsSL "$base/checksums.txt" -o "$cache/checksums.txt"; then
+        expected=""
+        while read -r checksum asset; do [[ $asset == "factory-ui-$os-$arch" ]] && expected=$checksum; done < "$cache/checksums.txt"
+        actual=""
+        if command -v shasum >/dev/null; then actual=$(shasum -a 256 "$binary.tmp"); actual=${actual%% *};
+        elif command -v sha256sum >/dev/null; then actual=$(sha256sum "$binary.tmp"); actual=${actual%% *}; fi
+        if [[ -n $expected && $actual == "$expected" ]] && gh attestation verify "$binary.tmp" --repo itsvedantkumar/factory-ai >/dev/null 2>&1; then mv "$binary.tmp" "$binary"; chmod 0755 "$binary"; else rm -f "$binary.tmp"; fi
+      fi
+    fi
+    if [[ -x $binary ]]; then
+      FACTORY_STORAGE_ACCOUNT="$STORAGE" FACTORY_NAME="$FACTORY_NAME" FACTORY_PURPOSE="$FACTORY_PURPOSE" exec "$binary"
+    fi
+    printf 'Charm client unavailable for this release; using Node fallback.\n' >&2
+    FACTORY_RESOURCE_GROUP="$RESOURCE_GROUP" FACTORY_VM="$VM" FACTORY_SERVICE_BUS="$NAMESPACE" FACTORY_KEY_VAULT="$VAULT" FACTORY_STORAGE_ACCOUNT="$STORAGE" FACTORY_NAME="$FACTORY_NAME" FACTORY_PURPOSE="$FACTORY_PURPOSE" node "$FACTORY_ROOT/src/tui.js"
     ;;
   setup)
     for dependency in az gh jq curl ssh-keygen node; do command -v "$dependency" >/dev/null || { printf 'Missing dependency: %s\n' "$dependency" >&2; exit 1; }; done
@@ -82,6 +204,8 @@ case "$command" in
     trap 'rm -f "$choices"' EXIT
     node "$FACTORY_ROOT/src/setup-menu.js" "$choices"
     provider=$(jq -r .provider "$choices")
+    factory_name=$(jq -r .factoryName "$choices")
+    factory_purpose=$(jq -r .factoryPurpose "$choices")
     location=$(jq -r .location "$choices")
     enterprise_org=$(jq -r .githubOrg "$choices")
     aws_region=$(jq -r .awsRegion "$choices")
@@ -107,6 +231,8 @@ case "$command" in
       printf 'FACTORY_SERVICE_BUS=%q\n' "$namespace"
       printf 'FACTORY_KEY_VAULT=%q\n' "$vault"
       printf 'FACTORY_STORAGE_ACCOUNT=%q\n' "$storage"
+      printf 'FACTORY_NAME=%q\n' "$factory_name"
+      printf 'FACTORY_PURPOSE=%q\n' "$factory_purpose"
     } > "$CONFIG_FILE"
     chmod 0600 "$CONFIG_FILE"
     ip=$(curl -fsS https://api.ipify.org)
@@ -154,7 +280,7 @@ case "$command" in
     printf 'Azure foundation and credential vault ready. Config: %s\n' "$CONFIG_FILE"
     if [[ $deploy_now == true ]]; then
       source_ref=$(gh api repos/itsvedantkumar/factory-ai/commits/main --jq .sha)
-      parameters=(KEY_VAULT_NAME "$vault" SERVICE_BUS_NAMESPACE "$namespace" SERVICE_BUS_QUEUE code-tasks FACTORY_STORAGE_ACCOUNT "$storage" SOURCE_REPOSITORY itsvedantkumar/factory-ai SOURCE_REF "$source_ref")
+      parameters=(KEY_VAULT_NAME "$vault" SERVICE_BUS_NAMESPACE "$namespace" FACTORY_STORAGE_ACCOUNT "$storage" FACTORY_NAME "$factory_name" FACTORY_PURPOSE "$factory_purpose" SOURCE_REPOSITORY itsvedantkumar/factory-ai SOURCE_REF "$source_ref")
       if [[ $provider == bedrock ]]; then
         parameters+=(AWS_REGION "$aws_region")
         for role in SCOUT PLANNER BUILDER TESTER DEBUGGER REVIEWER SECURITY RELEASE; do parameters+=("FACTORY_MODEL_$role" "bedrock/$bedrock_builder"); done
@@ -233,6 +359,7 @@ case "$command" in
     [[ -d $target ]] || { printf 'Project directory does not exist: %s\n' "$target" >&2; exit 2; }
     context="$target/.agent-factory"
     mkdir -p "$context"
+    [[ -e $target/AGENTS.md ]] || cp "$FACTORY_ROOT/templates/AGENTS.md" "$target/AGENTS.md"
     for template in "$FACTORY_ROOT/templates/project/"*.md; do
       destination="$context/$(basename "$template")"
       [[ -e $destination ]] || cp "$template" "$destination"

package/bootstrap/agent-factory-worker.service

@@ -1,8 +1,8 @@
 [Unit]
 Description=Factory AI durable worker
-After=network-online.target docker.service
-Wants=network-online.target
-Requires=docker.service
+After=network-online.target docker.service factory-ai-container-firewall.service factory-ai-qdrant.service factory-ai-ollama.service
+Wants=network-online.target factory-ai-qdrant.service factory-ai-ollama.service
+Requires=docker.service factory-ai-container-firewall.service
 StartLimitIntervalSec=300
 StartLimitBurst=10
 

package/bootstrap/auto-update.sh

@@ -18,6 +18,12 @@ set -a
 source /etc/agent-factory.env
 set +a
 github_token=$(az keyvault secret show --vault-name "$KEY_VAULT_NAME" --name github-token --query value --output tsv)
+tag_commit=$(GH_TOKEN="$github_token" gh api "repos/itsvedantkumar/factory-ai/commits/v$version" --jq .sha)
+[[ $tag_commit == "$commit" ]] || { printf 'npm gitHead does not match the immutable release tag.\n' >&2; exit 1; }
+release_ok=$(GH_TOKEN="$github_token" gh release view "v$version" --repo itsvedantkumar/factory-ai --json isDraft,isPrerelease --jq '(.isDraft == false) and (.isPrerelease == false)')
+[[ $release_ok == true ]] || { printf 'Matching stable GitHub release is unavailable.\n' >&2; exit 1; }
+provenance=$(npm view "factory-ai@$version" dist.attestations.provenance.url 2>/dev/null || true)
+[[ $provenance == https://search.sigstore.dev/* ]] || { printf 'npm release lacks verifiable provenance metadata.\n' >&2; exit 1; }
 checks=$(GH_TOKEN="$github_token" gh api "repos/itsvedantkumar/factory-ai/commits/$commit/check-runs" --jq '[.check_runs[] | select(.name == "verify") | .conclusion] | any(. == "success")')
 [[ $checks == true ]] || { printf 'Candidate commit lacks successful CI verification.\n' >&2; exit 1; }
 
@@ -41,20 +47,22 @@ docker run --rm --read-only --volume "$temporary/source:/workspace:ro" \
   detect --source /workspace --redact --no-banner --exit-code 1
 
 printf 'Deploying Factory AI %s (%s)\n' "$version" "$commit"
-if ! bash "$temporary/source/bootstrap/deploy-runtime.sh" \
-  KEY_VAULT_NAME "$KEY_VAULT_NAME" \
-  SERVICE_BUS_NAMESPACE "$SERVICE_BUS_NAMESPACE" \
-  SERVICE_BUS_QUEUE code-tasks \
-  SOURCE_REPOSITORY itsvedantkumar/factory-ai \
-  SOURCE_REF "$commit"; then
+deployment_parameters=(
+  KEY_VAULT_NAME "$KEY_VAULT_NAME"
+  SERVICE_BUS_NAMESPACE "$SERVICE_BUS_NAMESPACE"
+  SOURCE_REPOSITORY itsvedantkumar/factory-ai
+  SOURCE_REF "$commit"
+)
+for variable in FACTORY_STORAGE_ACCOUNT FACTORY_NAME FACTORY_PURPOSE AWS_REGION FACTORY_MODEL_SCOUT FACTORY_MODEL_PLANNER FACTORY_MODEL_BUILDER FACTORY_MODEL_TESTER FACTORY_MODEL_DEBUGGER FACTORY_MODEL_REVIEWER FACTORY_MODEL_SECURITY FACTORY_MODEL_RELEASE FACTORY_COMPACT_AFTER_INPUT_TOKENS FACTORY_COMPACT_MAX_CHARACTERS FACTORY_WATCHDOG_STALE_SECONDS FACTORY_HOOKS_JSON; do
+  [[ -n ${!variable:-} ]] && deployment_parameters+=("$variable" "${!variable}")
+done
+if ! bash "$temporary/source/bootstrap/deploy-runtime.sh" "${deployment_parameters[@]}"; then
   printf 'Candidate deployment failed. Restoring previous runtime.\n' >&2
   if [[ $previous_commit =~ ^[0-9a-f]{40}$ ]]; then
-    bash "$temporary/rollback-deploy.sh" \
-      KEY_VAULT_NAME "$KEY_VAULT_NAME" \
-      SERVICE_BUS_NAMESPACE "$SERVICE_BUS_NAMESPACE" \
-      SERVICE_BUS_QUEUE code-tasks \
-      SOURCE_REPOSITORY itsvedantkumar/factory-ai \
-      SOURCE_REF "$previous_commit"
+    for ((index=0; index<${#deployment_parameters[@]}; index+=2)); do
+      [[ ${deployment_parameters[index]} == SOURCE_REF ]] && deployment_parameters[index+1]=$previous_commit
+    done
+    bash "$temporary/rollback-deploy.sh" "${deployment_parameters[@]}"
   fi
   exit 1
 fi

package/bootstrap/deploy-runtime.sh

@@ -20,7 +20,7 @@ while (($# > 0)); do
     shift
   fi
   case "$name" in
-    KEY_VAULT_NAME|SERVICE_BUS_NAMESPACE|SERVICE_BUS_QUEUE|SOURCE_REPOSITORY|SOURCE_REF|FACTORY_STORAGE_ACCOUNT|AWS_REGION|FACTORY_MODEL_SCOUT|FACTORY_MODEL_PLANNER|FACTORY_MODEL_BUILDER|FACTORY_MODEL_TESTER|FACTORY_MODEL_DEBUGGER|FACTORY_MODEL_REVIEWER|FACTORY_MODEL_SECURITY|FACTORY_MODEL_RELEASE) printf -v "$name" '%s' "$value" ;;
+    KEY_VAULT_NAME|SERVICE_BUS_NAMESPACE|SOURCE_REPOSITORY|SOURCE_REF|FACTORY_STORAGE_ACCOUNT|FACTORY_NAME|FACTORY_PURPOSE|AWS_REGION|FACTORY_MODEL_SCOUT|FACTORY_MODEL_PLANNER|FACTORY_MODEL_BUILDER|FACTORY_MODEL_TESTER|FACTORY_MODEL_DEBUGGER|FACTORY_MODEL_REVIEWER|FACTORY_MODEL_SECURITY|FACTORY_MODEL_RELEASE|FACTORY_COMPACT_AFTER_INPUT_TOKENS|FACTORY_COMPACT_MAX_CHARACTERS|FACTORY_WATCHDOG_STALE_SECONDS|FACTORY_HOOKS_JSON) printf -v "$name" '%s' "$value" ;;
     *) echo "Unknown deployment parameter: $name" >&2; exit 1 ;;
   esac
 done
@@ -29,7 +29,6 @@ done
 : "${SERVICE_BUS_NAMESPACE:?SERVICE_BUS_NAMESPACE is required}"
 : "${SOURCE_REPOSITORY:?SOURCE_REPOSITORY as OWNER/REPOSITORY is required}"
 : "${SOURCE_REF:?SOURCE_REF must be a full commit SHA}"
-SERVICE_BUS_QUEUE=${SERVICE_BUS_QUEUE:-code-tasks}
 [[ $SOURCE_REPOSITORY =~ ^[A-Za-z0-9_.-]+/[A-Za-z0-9_.-]+$ ]] || { echo "Invalid SOURCE_REPOSITORY" >&2; exit 1; }
 [[ $SOURCE_REF =~ ^[0-9a-f]{40}$ ]] || { echo "SOURCE_REF must be a full lowercase commit SHA" >&2; exit 1; }
 
@@ -60,9 +59,10 @@ rsync -a --delete --exclude .git "$temporary/source/" /opt/agent-factory/app/
 
 KEY_VAULT_NAME="$KEY_VAULT_NAME" \
 SERVICE_BUS_NAMESPACE="$SERVICE_BUS_NAMESPACE" \
-SERVICE_BUS_QUEUE="$SERVICE_BUS_QUEUE" \
 FACTORY_WORKER_IMAGE="agent-factory-worker:$SOURCE_REF" \
 FACTORY_STORAGE_ACCOUNT="${FACTORY_STORAGE_ACCOUNT:-}" \
+FACTORY_NAME="${FACTORY_NAME:-Factory AI}" \
+FACTORY_PURPOSE="${FACTORY_PURPOSE:-Ship secure reviewed software continuously}" \
 AWS_REGION="${AWS_REGION:-}" \
 FACTORY_MODEL_SCOUT="${FACTORY_MODEL_SCOUT:-}" \
 FACTORY_MODEL_PLANNER="${FACTORY_MODEL_PLANNER:-}" \
@@ -72,6 +72,10 @@ FACTORY_MODEL_DEBUGGER="${FACTORY_MODEL_DEBUGGER:-}" \
 FACTORY_MODEL_REVIEWER="${FACTORY_MODEL_REVIEWER:-}" \
 FACTORY_MODEL_SECURITY="${FACTORY_MODEL_SECURITY:-}" \
 FACTORY_MODEL_RELEASE="${FACTORY_MODEL_RELEASE:-}" \
+FACTORY_COMPACT_AFTER_INPUT_TOKENS="${FACTORY_COMPACT_AFTER_INPUT_TOKENS:-80000}" \
+FACTORY_COMPACT_MAX_CHARACTERS="${FACTORY_COMPACT_MAX_CHARACTERS:-24000}" \
+FACTORY_WATCHDOG_STALE_SECONDS="${FACTORY_WATCHDOG_STALE_SECONDS:-900}" \
+FACTORY_HOOKS_JSON="${FACTORY_HOOKS_JSON:-[]}" \
 bash /opt/agent-factory/app/bootstrap/setup.sh
 install -d -o factory -g factory -m 0750 /opt/agent-factory/state
 version=$(node -p 'require("/opt/agent-factory/app/package.json").version')

package/bootstrap/factory-ai-container-firewall.service

@@ -0,0 +1,14 @@
+[Unit]
+Description=Block cloud metadata access from Factory AI containers
+After=docker.service
+Before=agent-factory-worker.service
+Requires=docker.service
+
+[Service]
+Type=oneshot
+ExecStart=/bin/bash -c '/usr/sbin/iptables -C DOCKER-USER -d 169.254.169.254/32 -j REJECT 2>/dev/null || /usr/sbin/iptables -I DOCKER-USER -d 169.254.169.254/32 -j REJECT'
+RemainAfterExit=yes
+NoNewPrivileges=true
+
+[Install]
+WantedBy=multi-user.target

package/bootstrap/factory-ai-ollama.service

@@ -0,0 +1,16 @@
+[Unit]
+Description=Factory AI local embedding inference
+After=docker.service
+Requires=docker.service
+
+[Service]
+Type=simple
+ExecStartPre=-/usr/bin/docker rm -f factory-ai-ollama
+ExecStart=/usr/bin/docker run --rm --name factory-ai-ollama --userns host --cap-drop ALL --security-opt no-new-privileges --pids-limit 1024 --memory 6g --cpus 4 --publish 127.0.0.1:11434:11434 --env OLLAMA_MODELS=/models --volume /opt/agent-factory/state/ollama:/models ollama/ollama@sha256:3d8a05e3432d50ea57594fabe971e46cc8fe963a0f9f8c40400bd56cd5388e47
+ExecStop=/usr/bin/docker stop -t 20 factory-ai-ollama
+Restart=always
+RestartSec=5
+SyslogIdentifier=factory-ai-ollama
+
+[Install]
+WantedBy=multi-user.target

package/bootstrap/factory-ai-qdrant.service

@@ -0,0 +1,16 @@
+[Unit]
+Description=Factory AI local Qdrant retrieval store
+After=docker.service
+Requires=docker.service
+
+[Service]
+Type=simple
+ExecStartPre=-/usr/bin/docker rm -f factory-ai-qdrant
+ExecStart=/usr/bin/docker run --rm --name factory-ai-qdrant --userns host --read-only --cap-drop ALL --security-opt no-new-privileges --pids-limit 1024 --memory 2g --cpus 2 --publish 127.0.0.1:6333:6333 --volume /opt/agent-factory/state/qdrant:/qdrant/storage --volume /opt/agent-factory/state/qdrant-snapshots:/qdrant/snapshots qdrant/qdrant@sha256:31407c0e8e32eb771b71718f1a4772e2ad47a07557917b21ac96792f40eb8007
+ExecStop=/usr/bin/docker stop -t 20 factory-ai-qdrant
+Restart=always
+RestartSec=5
+SyslogIdentifier=factory-ai-qdrant
+
+[Install]
+WantedBy=multi-user.target

package/bootstrap/factory-ai-snapshot.service

@@ -7,6 +7,7 @@ Wants=network-online.target
 Type=oneshot
 User=factory
 Group=factory
+SupplementaryGroups=systemd-journal
 WorkingDirectory=/opt/agent-factory/app
 EnvironmentFile=/etc/agent-factory-control.env
 ExecStart=/usr/bin/node /opt/agent-factory/app/src/snapshot.js

package/bootstrap/factory-ai-watchdog.service

@@ -0,0 +1,21 @@
+[Unit]
+Description=Recover stale Factory AI agent containers
+After=network-online.target docker.service agent-factory-control.service
+Wants=network-online.target
+Requires=docker.service
+
+[Service]
+Type=oneshot
+User=factory
+Group=factory
+SupplementaryGroups=docker
+WorkingDirectory=/opt/agent-factory/app
+EnvironmentFile=/etc/agent-factory.env
+ExecStart=/usr/bin/node /opt/agent-factory/app/src/watchdog.js
+NoNewPrivileges=true
+PrivateTmp=true
+ProtectSystem=strict
+ProtectHome=true
+ReadWritePaths=/opt/agent-factory/state /opt/agent-factory/workspaces
+UMask=0027
+SyslogIdentifier=factory-ai-watchdog

package/bootstrap/factory-ai-watchdog.timer

@@ -0,0 +1,11 @@
+[Unit]
+Description=Check Factory AI agent heartbeats every minute
+
+[Timer]
+OnBootSec=5m
+OnUnitActiveSec=1m
+RandomizedDelaySec=5
+Persistent=true
+
+[Install]
+WantedBy=timers.target

package/bootstrap/setup.sh

@@ -10,7 +10,6 @@ fi
 : "${KEY_VAULT_NAME:?Set KEY_VAULT_NAME to the existing vault name}"
 : "${SERVICE_BUS_NAMESPACE:?Set SERVICE_BUS_NAMESPACE to the existing namespace name}"
 : "${FACTORY_WORKER_IMAGE:?Set FACTORY_WORKER_IMAGE to an immutable image tag}"
-SERVICE_BUS_QUEUE=${SERVICE_BUS_QUEUE:-code-tasks}
 APP_DIR=${APP_DIR:-/opt/agent-factory/app}
 FACTORY_USER=${FACTORY_USER:-factory}
 
@@ -18,11 +17,11 @@ chmod 0755 /opt/agent-factory "$APP_DIR"
 
 export DEBIAN_FRONTEND=noninteractive
 apt-get update
-apt-get install -y ca-certificates curl gpg git jq
+apt-get install -y ca-certificates curl gpg git jq iptables
 
 install -d -m 0755 /etc/apt/keyrings
 curl -fsSL https://deb.nodesource.com/gpgkey/nodesource-repo.gpg.key | gpg --dearmor --yes -o /etc/apt/keyrings/nodesource.gpg
-printf '%s\n' 'deb [signed-by=/etc/apt/keyrings/nodesource.gpg] https://deb.nodesource.com/node_20.x nodistro main' > /etc/apt/sources.list.d/nodesource.list
+printf '%s\n' 'deb [signed-by=/etc/apt/keyrings/nodesource.gpg] https://deb.nodesource.com/node_22.x nodistro main' > /etc/apt/sources.list.d/nodesource.list
 curl -fsSL https://cli.github.com/packages/githubcli-archive-keyring.gpg -o /etc/apt/keyrings/githubcli-archive-keyring.gpg
 chmod go+r /etc/apt/keyrings/githubcli-archive-keyring.gpg
 printf '%s\n' "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" > /etc/apt/sources.list.d/github-cli.list
@@ -30,7 +29,7 @@ curl -fsSL https://packages.microsoft.com/keys/microsoft.asc | gpg --dearmor --y
 printf '%s\n' "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/microsoft.gpg] https://packages.microsoft.com/repos/azure-cli/ $(. /etc/os-release && echo "$VERSION_CODENAME") main" > /etc/apt/sources.list.d/azure-cli.list
 apt-get update
 apt-get install -y azure-cli gh nodejs
-test "$(node --version | cut -d. -f1)" = "v20"
+test "$(node --version | cut -d. -f1)" = "v22"
 
 id "$FACTORY_USER" >/dev/null 2>&1 || useradd --system --create-home --shell /usr/sbin/nologin "$FACTORY_USER"
 test -f "$APP_DIR/package-lock.json"
@@ -44,6 +43,9 @@ for scanner_image in \
   semgrep/semgrep@sha256:183a149fb3e9700ab5294a7b4ab0241a826fd046bc8b721062fbea80fdfa438f; do
   docker pull "$scanner_image"
 done
+docker pull ollama/ollama@sha256:3d8a05e3432d50ea57594fabe971e46cc8fe963a0f9f8c40400bd56cd5388e47
+docker pull qdrant/qdrant@sha256:31407c0e8e32eb771b71718f1a4772e2ad47a07557917b21ac96792f40eb8007
+iptables -C DOCKER-USER -d 169.254.169.254/32 -j REJECT 2>/dev/null || iptables -I DOCKER-USER -d 169.254.169.254/32 -j REJECT
 
 STATE_DEVICE=/dev/disk/azure/scsi1/lun0
 if [[ -b $STATE_DEVICE ]]; then
@@ -65,7 +67,11 @@ for secret_name in "${GITHUB_TOKEN_SECRET:-github-token}"; do
   az keyvault secret show --vault-name "$KEY_VAULT_NAME" --name "$secret_name" --query id --output none
 done
 
-install -d -o "$FACTORY_USER" -g "$FACTORY_USER" -m 0750 /opt/agent-factory/state /opt/agent-factory/state/home /opt/agent-factory/state/memory /opt/agent-factory/state/telegram /opt/agent-factory/workspaces /opt/agent-factory/logs
+install -d -o "$FACTORY_USER" -g "$FACTORY_USER" -m 0750 /opt/agent-factory/state /opt/agent-factory/state/home /opt/agent-factory/state/memory /opt/agent-factory/state/telegram /opt/agent-factory/state/retrieval /opt/agent-factory/workspaces /opt/agent-factory/logs
+if [[ -f /opt/agent-factory/state/memory/knowledge-graph.jsonl && ! -e /opt/agent-factory/state/memory/legacy-unscoped-knowledge-graph.jsonl ]]; then
+  mv /opt/agent-factory/state/memory/knowledge-graph.jsonl /opt/agent-factory/state/memory/legacy-unscoped-knowledge-graph.jsonl
+fi
+install -d -o root -g root -m 0750 /opt/agent-factory/state/qdrant /opt/agent-factory/state/qdrant-snapshots /opt/agent-factory/state/ollama
 install -m 0600 -o root -g root /dev/null /etc/agent-factory.env
 {
   printf 'SERVICE_BUS_NAMESPACE=%s\n' "$SERVICE_BUS_NAMESPACE"
@@ -81,10 +87,15 @@ install -m 0600 -o root -g root /dev/null /etc/agent-factory.env
   printf 'FACTORY_REGISTRY=%s/config/capabilities.json\n' "$APP_DIR"
   printf 'FACTORY_WORKER_IMAGE=%s\n' "$FACTORY_WORKER_IMAGE"
   printf 'FACTORY_VERSION=%s\n' "$factory_version"
+  printf 'FACTORY_NAME=%s\n' "${FACTORY_NAME:-Factory AI}"
+  printf 'FACTORY_PURPOSE=%s\n' "${FACTORY_PURPOSE:-Ship secure reviewed software continuously}"
   printf 'MAX_CONCURRENCY=3\n'
   printf 'TASK_TIMEOUT_MS=1800000\n'
+  printf 'FACTORY_COMPACT_AFTER_INPUT_TOKENS=%s\n' "${FACTORY_COMPACT_AFTER_INPUT_TOKENS:-80000}"
+  printf 'FACTORY_COMPACT_MAX_CHARACTERS=%s\n' "${FACTORY_COMPACT_MAX_CHARACTERS:-24000}"
+  printf 'FACTORY_WATCHDOG_STALE_SECONDS=%s\n' "${FACTORY_WATCHDOG_STALE_SECONDS:-900}"
+  printf 'FACTORY_HOOKS_JSON=%s\n' "${FACTORY_HOOKS_JSON:-[]}"
   printf 'MAX_DELIVERY_COUNT=8\n'
-  printf 'FACTORY_VERSION=%s\n' "$factory_version"
   [[ -n ${AWS_REGION:-} ]] && printf 'AWS_REGION=%s\nAWS_DEFAULT_REGION=%s\n' "$AWS_REGION" "$AWS_REGION"
   for variable in FACTORY_MODEL_SCOUT FACTORY_MODEL_PLANNER FACTORY_MODEL_BUILDER FACTORY_MODEL_TESTER FACTORY_MODEL_DEBUGGER FACTORY_MODEL_REVIEWER FACTORY_MODEL_SECURITY FACTORY_MODEL_RELEASE; do
     [[ -n ${!variable:-} ]] && printf '%s=%s\n' "$variable" "${!variable}"
@@ -104,6 +115,8 @@ install -m 0600 -o root -g root /dev/null /etc/agent-factory-control.env
   printf 'FACTORY_STATE_DIR=/opt/agent-factory/state\n'
   printf 'FACTORY_REGISTRY=%s/config/capabilities.json\n' "$APP_DIR"
   printf 'MAX_DELIVERY_COUNT=8\n'
+  printf 'FACTORY_COMPACT_AFTER_INPUT_TOKENS=%s\n' "${FACTORY_COMPACT_AFTER_INPUT_TOKENS:-80000}"
+  printf 'FACTORY_COMPACT_MAX_CHARACTERS=%s\n' "${FACTORY_COMPACT_MAX_CHARACTERS:-24000}"
   [[ -n ${AWS_REGION:-} ]] && printf 'AWS_REGION=%s\nAWS_DEFAULT_REGION=%s\n' "$AWS_REGION" "$AWS_REGION"
   for variable in FACTORY_MODEL_SCOUT FACTORY_MODEL_PLANNER FACTORY_MODEL_BUILDER FACTORY_MODEL_TESTER FACTORY_MODEL_DEBUGGER FACTORY_MODEL_REVIEWER FACTORY_MODEL_SECURITY FACTORY_MODEL_RELEASE; do
     [[ -n ${!variable:-} ]] && printf '%s=%s\n' "$variable" "${!variable}"
@@ -111,9 +124,12 @@ install -m 0600 -o root -g root /dev/null /etc/agent-factory-control.env
 } > /etc/agent-factory-control.env
 
 chown -R "$FACTORY_USER:$FACTORY_USER" /opt/agent-factory/state /opt/agent-factory/workspaces /opt/agent-factory/logs
+chown root:root /opt/agent-factory/state/qdrant /opt/agent-factory/state/qdrant-snapshots /opt/agent-factory/state/ollama
+chmod 0750 /opt/agent-factory/state/qdrant /opt/agent-factory/state/qdrant-snapshots /opt/agent-factory/state/ollama
 chown -R root:root "$APP_DIR"
 chmod -R go-w "$APP_DIR"
 install -m 0644 "$APP_DIR/bootstrap/agent-factory-worker.service" /etc/systemd/system/agent-factory-worker.service
+install -m 0644 "$APP_DIR/bootstrap/factory-ai-container-firewall.service" /etc/systemd/system/factory-ai-container-firewall.service
 install -m 0644 "$APP_DIR/bootstrap/agent-factory-control.service" /etc/systemd/system/agent-factory-control.service
 install -m 0644 "$APP_DIR/bootstrap/agent-factory-release.service" /etc/systemd/system/agent-factory-release.service
 install -m 0644 "$APP_DIR/bootstrap/agent-factory-reporter.service" /etc/systemd/system/agent-factory-reporter.service
@@ -123,14 +139,23 @@ chmod 0755 "$APP_DIR/bootstrap/auto-update.sh"
 install -m 0644 "$APP_DIR/bootstrap/factory-ai-update.service" /etc/systemd/system/factory-ai-update.service
 install -m 0644 "$APP_DIR/bootstrap/factory-ai-update.timer" /etc/systemd/system/factory-ai-update.timer
 install -m 0644 "$APP_DIR/bootstrap/factory-ai-snapshot.service" /etc/systemd/system/factory-ai-snapshot.service
-install -m 0644 "$APP_DIR/bootstrap/factory-ai-snapshot.timer" /etc/systemd/system/factory-ai-snapshot.timer
+  install -m 0644 "$APP_DIR/bootstrap/factory-ai-snapshot.timer" /etc/systemd/system/factory-ai-snapshot.timer
+  install -m 0644 "$APP_DIR/bootstrap/factory-ai-watchdog.service" /etc/systemd/system/factory-ai-watchdog.service
+  install -m 0644 "$APP_DIR/bootstrap/factory-ai-watchdog.timer" /etc/systemd/system/factory-ai-watchdog.timer
+install -m 0644 "$APP_DIR/bootstrap/factory-ai-qdrant.service" /etc/systemd/system/factory-ai-qdrant.service
+install -m 0644 "$APP_DIR/bootstrap/factory-ai-ollama.service" /etc/systemd/system/factory-ai-ollama.service
 systemctl daemon-reload
+systemctl enable --now factory-ai-container-firewall.service
 systemctl enable --now agent-factory-worker.service
 systemctl enable --now agent-factory-control.service
 systemctl enable --now agent-factory-release.service
 systemctl enable --now agent-factory-reporter.timer
 systemctl enable --now agent-factory-telegram.service
 systemctl enable --now factory-ai-update.timer
-systemctl enable --now factory-ai-snapshot.timer
+  systemctl enable --now factory-ai-snapshot.timer
+  systemctl enable --now factory-ai-watchdog.timer
+systemctl enable --now factory-ai-qdrant.service factory-ai-ollama.service
+for _ in $(seq 1 60); do curl -fsS http://127.0.0.1:11434/api/tags >/dev/null && break; sleep 2; done
+docker exec factory-ai-ollama ollama show embeddinggemma >/dev/null 2>&1 || docker exec factory-ai-ollama ollama pull embeddinggemma
 systemctl restart agent-factory-control.service agent-factory-worker.service agent-factory-release.service
 echo "Agent factory worker installed"