tomo-ai 0.8.10 → 0.8.12

package/CHANGELOG.md

@@ -1,5 +1,29 @@
 # Changelog
 
+## 0.8.12 (2026-07-12)
+
+### Features
+
+- **Prometheus metrics + activity log with Grafana stack** (#238). Two new `WatchBus` subscribers, gated by `config.metrics` (default off): a `prom-client` exporter on `127.0.0.1:9464` (turns, cost, per-session context, tool calls, cron, heartbeats, compactions) and an NDJSON activity log at `~/.tomo/logs/activity.ndjson` for Loki tailing (transcript text local-only, redactable via `metrics.includeMessageText=false`). `contrib/observability/` ships a docker-compose stack (Prometheus, Loki, Alloy, Grafana) with a provisioned Tomo dashboard.
+
+### Dependencies
+
+- Bump `@anthropic-ai/claude-agent-sdk` 0.3.201 → 0.3.207 (#237).
+- Bump the dev-dependencies group (#236). The TypeScript ^6 → ^7 bump in that group was reverted — no `typescript-eslint` release yet supports TypeScript 7 (peer range caps at `<6.1.0`), so it broke `npm ci`.
+
+## 0.8.11 (2026-07-09)
+
+### Features
+
+- **`imsg`-backed iMessage channel (BlueBubbles successor)** (#230, #231, #232, #234). A new `ImsgChannel` (`src/channels/imessage-imsg.ts`) implements the `Channel` surface over a single long-lived `imsg rpc` child (newline-delimited JSON-RPC 2.0 on stdio), selectable via `channels.imessage.provider: "imsg"` without touching the BlueBubbles channel. Inbound uses the `watch.subscribe` all-chat stream with attachments (local file paths), reply context, inbound tapbacks, slash commands, and group/DM mapping; outbound covers chunked sends, threaded replies via `send.rich` (with plain-send fallback), attachments, targeted tapbacks, typing indicators, mark-read, and unsend. Message edit is gated on the `imsg status --json` bridge selector probe, so on macOS 26 (where the edit selectors are gone OS-wide) the channel refuses cleanly instead of crashing Messages. The child restarts on crash with escalating backoff and resumes the watch from a persisted rowid cursor, and persistent GUID dedupe is shared with the BlueBubbles store so a provider cutover never re-dispatches messages. Session keys stay compatible — chat.db chat GUIDs are used verbatim (identical to what BlueBubbles reported), so existing session mappings survive. Inbound HEIC/HEIF attachments that `imsg` doesn't pre-convert are normalized to JPEG channel-side via macOS `sips` (by mime, `.heic`/`.heif` extension, or `ftyp` magic-byte sniff), read-only-safe (converts into a temp JPEG that is read then unlinked) and never dropped on conversion failure.
+- **`tomo watch` — live mission-control TUI** (#229). An always-on terminal dashboard for watching what Tomo is doing, backed by a daemon-side event bus. `src/watch/bus.ts` is an in-process `WatchBus` singleton with a ring buffer, fed by emit hooks in `SessionStore.append` (inbound/outbound messages), `LiveSession` (tool start/end with durations, compact boundaries), `TurnRunner` (turn start/end with ingress source), the per-turn cost/context stats closure, `CronScheduler`, `ContinuityRunner`, and a pino tap (warn/error → "issue" events). `server.ts` serves NDJSON over `~/.tomo/watch.sock` (0600) — a snapshot on connect (vitals, cron next-runs, cost 24h/7d, feed backfill), then live relay; inbound "send" frames route chat into the dm session. The Ink/React TUI renders an activity feed (messages, tools with status + duration, turn summaries with cost + ctx%, cron, heartbeats), a pinned in-flight turn line, a vitals sidebar (context gauge, cost, next-up, last warn/error), a chat line, a raw-log-tail toggle, a group-traffic filter, scroll with follow mode, and a help overlay. Auto-reconnects with an offline header state and drops to a 15s idle render tick so it can stay open 24/7. The daemon never depends on clients.
+
+### Bug fixes
+
+- **Trailing `NO_REPLY` suppresses the whole block/turn** (#233). A reply whose final line(s) are a bare `NO_REPLY` is now suppressed entirely instead of delivering the stripped remainder — the agent narrates housekeeping turns and ends them with `NO_REPLY`, and that narration (sometimes including the literal `NO_REPLY` text) previously leaked to the channel (owner decision 2026-07-08: trailing `NO_REPLY` marks the whole response as not-for-the-channel). `agent/text-utils` gains `endsWithTrailingNoReply()` (`stripTrailingNoReply` stays exported for the watch TUI, which still shows the visible text); `TurnRunner.runSendTurn` (cron/continuity) silences the entire turn; imsg and BlueBubbles streaming drop a block whose trailing line is bare `NO_REPLY`; and Telegram streaming retracts the already-streamed message before the multi-part delivery path. Follow-up rounds close the Telegram over-4096 rollover gaps: finalized rollover head chunks are now deleted and un-recorded (new `forgetOwnMessage`) on every block-abandoning path (`finalFlush` suppression, multi-part re-delivery, `discardBlock`, `cancel`), suppression is judged on the full block buffer rather than the post-rollover slice, and attachment blocks (`MEDIA:`/`STICKER:`) are suppressed as a whole.
+- **iMessage inbound reply context only for genuine threaded replies** (#234). Since the imsg cutover, nearly every inbound iMessage arrived tagged `[replying to: "..."]` even when the sender never used long-press → Reply, because the channel fell back to `reply_to_guid` — a chat.db column populated on almost every row (it points at the preceding message, usually our own last outbound; verified 186/186 inbound DM rows over 3 days carried it, vs. 7 with a genuine thread originator). The reply marker is now gated on `thread_originator_guid` only, with fallback to the recent-message ring and then a quote-less marker. Outbound threading and tapback handling are untouched.
+- **imsg channel post-merge review fixes** (#232). `stop()` no longer leaves in-flight requests hanging forever — `killChild()` now rejects pending requests before nulling the child. A failed row with no numeric rowid (parsed as `0`) no longer wedges the cursor floor permanently — it is logged and skipped. A write parked on `drain` now settles when its response arrives (the response proves the child consumed the line), so a healthy answering child is never killed by the drain backstop. A temp JPEG from a failed HEIC conversion is unlinked instead of leaking.
+
 ## 0.8.10 (2026-07-06)
 
 ### Features

package/README.md

@@ -108,11 +108,12 @@ Tomo keeps person records at `~/.tomo/workspace/memory/people/*.md` — canonica
   - Image/photo support (sends to Claude as vision input)
   - Group chat: defaults to mention-required (only responds when @mentioned or replied to); per-group passive listen mode opt-in via `channels.telegram.passiveGroups`
   - Markdown rendering with plain-text fallback
-- **iMessage** — via [BlueBubbles](https://bluebubbles.app)
+- **iMessage** — via [BlueBubbles](https://bluebubbles.app) (default) or the [imsg CLI](https://github.com/openclaw/imsg) (`channels.imessage.provider: "imsg"`)
   - DM and group chat support
   - Image attachment support
   - Contact name resolution from Mac contacts
   - Group chat: observes all messages, only responds when relevant (replies `NO_REPLY` to stay silent)
+  - imsg provider: no external server — Tomo spawns one `imsg rpc` child (needs Full Disk Access); adds inbound tapbacks and built-in reply context, and gates message editing on a live macOS selector probe
 
 ### Multi-Channel Sessions
 
@@ -291,6 +292,16 @@ Run `tomo config` for interactive setup, or edit `~/.tomo/config.json` directly:
 }
 ```
 
+To run iMessage through the local [imsg CLI](https://github.com/openclaw/imsg) instead of a BlueBubbles server (the Tomo process needs Full Disk Access; advanced features additionally need `imsg launch`):
+
+```json
+{
+  "channels": {
+    "imessage": { "provider": "imsg", "cliPath": "/opt/homebrew/bin/imsg", "allowlist": ["+15551234567"] }
+  }
+}
+```
+
 Environment variables override config file values:
 
 | Variable | Description |
@@ -298,6 +309,9 @@ Environment variables override config file values:
 | `ANTHROPIC_API_KEY` | Use Anthropic API-key authentication; overrides the authentication stored in `config.json` |
 | `TELEGRAM_BOT_TOKEN` | Override Telegram token |
 | `IMESSAGE_URL` | Override BlueBubbles URL |
+| `IMESSAGE_PROVIDER` | iMessage backend: `bluebubbles` (default) or `imsg` |
+| `IMSG_CLI_PATH` | Path to the `imsg` binary (default: `imsg` on PATH) |
+| `IMSG_DB_PATH` | Override chat.db path passed to `imsg rpc --db` |
 | `IMESSAGE_TYPING_START_DELAY_MS` | Delay before showing iMessage typing for ordinary turns (default: `1200`) |
 | `IMESSAGE_PASSIVE_TYPING_START_DELAY_MS` | Delay before showing iMessage typing for passive iMessage group turns (default: `4000`) |
 | `CLAUDE_MODEL` | Override model |

package/contrib/observability/README.md

@@ -0,0 +1,90 @@
+# Tomo observability stack
+
+A self-contained Grafana + Prometheus + Loki + Alloy stack for watching Tomo
+from a dashboard. Data never leaves the machine: Prometheus, Loki, and Alloy
+bind to `127.0.0.1` only. Grafana is reachable from the local network (login
+required) so you can open the dashboard from another computer.
+
+```
+Tomo daemon ──/metrics (127.0.0.1:9464)──▶ Prometheus ──┐
+     │                                                  ├──▶ Grafana (127.0.0.1:3000)
+     └──~/.tomo/logs/activity.ndjson──▶ Alloy ──▶ Loki ─┘
+```
+
+## Setup
+
+1. Enable metrics in `~/.tomo/config.json`:
+
+   ```json
+   {
+     "metrics": { "enabled": true }
+   }
+   ```
+
+   Then restart the daemon (`tomo restart`) and verify:
+
+   ```bash
+   curl -s localhost:9464/metrics | grep tomo_
+   ```
+
+2. Copy the stack somewhere you own, then start it (requires Docker Desktop).
+   Copying matters: run it in place and an `npm upgrade` will overwrite any
+   compose/dashboard customizations.
+
+   ```bash
+   # from a git clone:
+   cp -r contrib/observability ~/tomo-observability
+   # or from a global npm install:
+   cp -r "$(npm root -g)/tomo-ai/contrib/observability" ~/tomo-observability
+
+   cd ~/tomo-observability
+   docker compose up -d
+   ```
+
+3. Open <http://localhost:3000> — or from another computer on your network,
+   `http://<hostname>.local:3000` (e.g. `http://mac-mini.local:3000`). Sign in
+   with `admin` / `admin`; Grafana forces a password change on first login.
+   The **Tomo** dashboard is pre-provisioned.
+
+   To restrict Grafana to this machine only, change its port mapping back to
+   `"127.0.0.1:3000:3000"` in `docker-compose.yml`. For access from *outside*
+   your home network, don't port-forward — use Tailscale or similar.
+
+## Config reference
+
+`~/.tomo/config.json` → `metrics` (env overrides in parens):
+
+| Key | Default | Meaning |
+| --- | --- | --- |
+| `enabled` (`TOMO_METRICS`) | `false` | Serve `/metrics` and write the activity log |
+| `port` (`TOMO_METRICS_PORT`) | `9464` | Prometheus exporter port, loopback-only |
+| `activityLog` | `true` | Write `~/.tomo/logs/activity.ndjson` for Loki |
+| `includeMessageText` | `true` | Include transcript text in the activity log. Disable if you ever ship the log off this machine. |
+
+## What's collected
+
+**Prometheus** (`tomo_*`, plus standard `process_*`/`nodejs_*`): turns by
+source/outcome, turn duration histogram, cumulative cost (USD), per-session
+context usage, tool calls, cron runs + upcoming cron schedule, compactions,
+heartbeats (count, last, next), warn/error issues.
+
+**Loki** (`{job="tomo-activity"}`): every watch-bus event as structured JSON —
+the same feed `tomo watch` renders. Labels: `type`, `session`, `level`.
+Example queries:
+
+```logql
+{job="tomo-activity", type="tool.start"}                # every tool call
+{job="tomo-activity", type="transcript", session="dm:merlin"}
+{job="tomo-activity"} | json | costUsd > 0.10           # expensive turns
+```
+
+## Notes
+
+- Prometheus keeps 90 days of metrics; Loki uses its default retention.
+  Volumes (`prometheus-data`, `loki-data`, `grafana-data`) survive
+  `docker compose down`; add `-v` to wipe them.
+- Counters reset when the daemon restarts — dashboard queries use
+  `increase()`/`rate()`, which handle resets.
+- If `TOMO_HOME` is not `~/.tomo`, fix the Alloy volume mount in
+  `docker-compose.yml`.
+- History starts when you enable the stack; nothing is backfilled.

package/contrib/observability/alloy/config.alloy

@@ -0,0 +1,53 @@
+// Tail Tomo's NDJSON activity feed into Loki. The glob also matches rotated
+// files (activity.ndjson.1, …) so lines written just before a rotation are
+// still picked up if the shipper lags.
+
+local.file_match "tomo_activity" {
+	path_targets = [{ __path__ = "/var/log/tomo/activity.ndjson*" }]
+}
+
+loki.source.file "tomo_activity" {
+	targets    = local.file_match.tomo_activity.targets
+	forward_to = [loki.process.tomo.receiver]
+}
+
+loki.process "tomo" {
+	stage.json {
+		expressions = {
+			type    = "type",
+			session = "sessionKey",
+			level   = "level",
+			ts      = "ts",
+		}
+	}
+
+	// Low-cardinality labels only: event type, session key, issue level.
+	// Everything else stays in the JSON body and is queryable via `| json`.
+	stage.labels {
+		values = {
+			type    = "",
+			session = "",
+			level   = "",
+		}
+	}
+
+	stage.static_labels {
+		values = {
+			job = "tomo-activity",
+		}
+	}
+
+	// Use the event's own millisecond timestamp, not ingestion time.
+	stage.timestamp {
+		source = "ts"
+		format = "UnixMs"
+	}
+
+	forward_to = [loki.write.local.receiver]
+}
+
+loki.write "local" {
+	endpoint {
+		url = "http://loki:3100/loki/api/v1/push"
+	}
+}

package/contrib/observability/docker-compose.yml

@@ -0,0 +1,70 @@
+# Local observability stack for Tomo: Prometheus (metrics), Loki (activity
+# feed), Alloy (log shipper), Grafana (dashboard). See README.md.
+#
+# Grafana listens on the LAN (with login required); everything else binds to
+# 127.0.0.1 — Prometheus/Loki have no auth of their own, and Grafana reaches
+# them over the internal Docker network, so they never need outside exposure.
+services:
+  prometheus:
+    image: prom/prometheus:v3.4.1
+    ports:
+      - "127.0.0.1:9090:9090"
+    volumes:
+      - ./prometheus.yml:/etc/prometheus/prometheus.yml:ro
+      - prometheus-data:/prometheus
+    command:
+      - --config.file=/etc/prometheus/prometheus.yml
+      - --storage.tsdb.retention.time=90d
+    extra_hosts:
+      # Docker Desktop for Mac provides host.docker.internal natively; this
+      # line makes the same name work if the stack ever runs on Linux.
+      - "host.docker.internal:host-gateway"
+    restart: unless-stopped
+
+  loki:
+    image: grafana/loki:3.5.1
+    # No host port: Grafana and Alloy reach Loki over the internal Docker
+    # network, and host 3100 is BlueBubbles' default webhook port — publishing
+    # it would collide with a default iMessage setup. To poke Loki directly:
+    #   docker compose exec loki wget -qO- localhost:3100/ready
+    volumes:
+      - loki-data:/loki
+    restart: unless-stopped
+
+  alloy:
+    image: grafana/alloy:v1.9.1
+    volumes:
+      - ./alloy/config.alloy:/etc/alloy/config.alloy:ro
+      - alloy-data:/var/lib/alloy/data
+      # Tomo's activity log (metrics.activityLog). Adjust if TOMO_HOME moved.
+      - ~/.tomo/logs:/var/log/tomo:ro
+    command:
+      - run
+      - /etc/alloy/config.alloy
+      - --storage.path=/var/lib/alloy/data
+    depends_on:
+      - loki
+    restart: unless-stopped
+
+  grafana:
+    image: grafana/grafana:12.0.2
+    ports:
+      # LAN-reachable so other machines can open the dashboard. Because the
+      # activity feed can contain message text, login stays REQUIRED — first
+      # sign-in is admin/admin and Grafana forces a password change. To lock
+      # it back down to this machine only, prefix with 127.0.0.1: again.
+      - "3000:3000"
+    volumes:
+      - ./grafana/provisioning:/etc/grafana/provisioning:ro
+      - ./grafana/dashboards:/var/lib/grafana/dashboards:ro
+      - grafana-data:/var/lib/grafana
+    depends_on:
+      - prometheus
+      - loki
+    restart: unless-stopped
+
+volumes:
+  prometheus-data:
+  loki-data:
+  alloy-data:
+  grafana-data:

package/contrib/observability/grafana/dashboards/tomo.json

@@ -0,0 +1,328 @@
+{
+  "uid": "tomo-overview",
+  "title": "Tomo",
+  "tags": ["tomo"],
+  "schemaVersion": 39,
+  "editable": true,
+  "time": { "from": "now-24h", "to": "now" },
+  "refresh": "30s",
+  "panels": [
+    {
+      "id": 1,
+      "type": "stat",
+      "title": "Daemon",
+      "gridPos": { "h": 4, "w": 3, "x": 0, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "up{job=\"tomo\"}" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] }, "colorMode": "background" },
+      "fieldConfig": {
+        "defaults": {
+          "mappings": [
+            {
+              "type": "value",
+              "options": {
+                "0": { "text": "DOWN", "color": "red" },
+                "1": { "text": "UP", "color": "green" }
+              }
+            }
+          ]
+        },
+        "overrides": []
+      }
+    },
+    {
+      "id": 2,
+      "type": "stat",
+      "title": "Cost (24h)",
+      "gridPos": { "h": 4, "w": 3, "x": 3, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "sum(increase(tomo_cost_usd_total[24h]))" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": { "defaults": { "unit": "currencyUSD", "decimals": 2 }, "overrides": [] }
+    },
+    {
+      "id": 3,
+      "type": "stat",
+      "title": "Turns (24h)",
+      "gridPos": { "h": 4, "w": 3, "x": 6, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "sum(increase(tomo_turns_total[24h]))" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": { "defaults": { "unit": "short", "decimals": 0 }, "overrides": [] }
+    },
+    {
+      "id": 4,
+      "type": "stat",
+      "title": "Errors (24h)",
+      "gridPos": { "h": 4, "w": 3, "x": 9, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        { "refId": "A", "expr": "sum(increase(tomo_issues_total{level=\"error\"}[24h])) or vector(0)" }
+      ],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "short",
+          "decimals": 0,
+          "thresholds": {
+            "mode": "absolute",
+            "steps": [
+              { "color": "green", "value": null },
+              { "color": "red", "value": 1 }
+            ]
+          }
+        },
+        "overrides": []
+      }
+    },
+    {
+      "id": 5,
+      "type": "stat",
+      "title": "In flight",
+      "gridPos": { "h": 4, "w": 3, "x": 12, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "tomo_turns_active" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": { "defaults": { "unit": "short", "decimals": 0 }, "overrides": [] }
+    },
+    {
+      "id": 6,
+      "type": "stat",
+      "title": "Memory (RSS)",
+      "gridPos": { "h": 4, "w": 3, "x": 15, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "process_resident_memory_bytes{job=\"tomo\"}" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": { "defaults": { "unit": "bytes" }, "overrides": [] }
+    },
+    {
+      "id": 7,
+      "type": "stat",
+      "title": "Next heartbeat",
+      "gridPos": { "h": 4, "w": 3, "x": 18, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "tomo_heartbeat_next_timestamp_seconds * 1000" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": {
+        "defaults": {
+          "unit": "dateTimeFromNow",
+          "noValue": "n/a",
+          "mappings": [
+            {
+              "type": "value",
+              "options": { "0": { "text": "continuity off", "color": "text" } }
+            }
+          ]
+        },
+        "overrides": []
+      }
+    },
+    {
+      "id": 8,
+      "type": "stat",
+      "title": "Next cron job",
+      "gridPos": { "h": 4, "w": 3, "x": 21, "y": 0 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [{ "refId": "A", "expr": "min(tomo_cron_next_run_timestamp_seconds) * 1000" }],
+      "options": { "reduceOptions": { "calcs": ["lastNotNull"] } },
+      "fieldConfig": { "defaults": { "unit": "dateTimeFromNow", "noValue": "none" }, "overrides": [] }
+    },
+    {
+      "id": 21,
+      "type": "table",
+      "title": "Upcoming cron jobs",
+      "gridPos": { "h": 9, "w": 8, "x": 0, "y": 4 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "tomo_cron_next_run_timestamp_seconds * 1000",
+          "format": "table",
+          "instant": true
+        }
+      ],
+      "transformations": [
+        {
+          "id": "organize",
+          "options": {
+            "excludeByName": { "Time": true, "__name__": true, "instance": true, "job": true },
+            "renameByName": { "cron_job": "Job", "Value": "Next run" }
+          }
+        }
+      ],
+      "options": { "sortBy": [{ "displayName": "Next run", "desc": false }] },
+      "fieldConfig": {
+        "defaults": { "unit": "dateTimeFromNow" },
+        "overrides": [
+          {
+            "matcher": { "id": "byName", "options": "Job" },
+            "properties": [{ "id": "unit", "value": "string" }]
+          }
+        ]
+      }
+    },
+    {
+      "id": 22,
+      "type": "logs",
+      "title": "Tool calls",
+      "gridPos": { "h": 9, "w": 16, "x": 8, "y": 4 },
+      "datasource": { "type": "loki", "uid": "loki" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "{job=\"tomo-activity\", type=\"tool.start\"} | json | line_format \"{{.tool}}{{if .agent}} [{{.agent}}]{{end}}{{if .detail}} — {{.detail}}{{end}}\""
+        },
+        {
+          "refId": "B",
+          "expr": "{job=\"tomo-activity\", type=\"tool.end\"} | json | ok=\"false\" | line_format \"{{.tool}} FAILED{{if .durationMs}} after {{.durationMs}}ms{{end}}\""
+        }
+      ],
+      "options": {
+        "showTime": true,
+        "sortOrder": "Descending",
+        "wrapLogMessage": true,
+        "enableLogDetails": true,
+        "dedupStrategy": "none"
+      }
+    },
+    {
+      "id": 10,
+      "type": "timeseries",
+      "title": "Cost per hour",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 13 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        { "refId": "A", "expr": "sum(increase(tomo_cost_usd_total[1h]))", "legendFormat": "USD/h" }
+      ],
+      "fieldConfig": { "defaults": { "unit": "currencyUSD" }, "overrides": [] }
+    },
+    {
+      "id": 11,
+      "type": "timeseries",
+      "title": "Turns per hour by source",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 13 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (source) (increase(tomo_turns_total[1h]))",
+          "legendFormat": "{{source}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": { "unit": "short", "custom": { "stacking": { "mode": "normal" } } },
+        "overrides": []
+      }
+    },
+    {
+      "id": 12,
+      "type": "timeseries",
+      "title": "Turn duration (p50 / p95)",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 21 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "histogram_quantile(0.5, sum by (le) (rate(tomo_turn_duration_seconds_bucket[$__rate_interval])))",
+          "legendFormat": "p50"
+        },
+        {
+          "refId": "B",
+          "expr": "histogram_quantile(0.95, sum by (le) (rate(tomo_turn_duration_seconds_bucket[$__rate_interval])))",
+          "legendFormat": "p95"
+        }
+      ],
+      "fieldConfig": { "defaults": { "unit": "s" }, "overrides": [] }
+    },
+    {
+      "id": 13,
+      "type": "timeseries",
+      "title": "Context usage by session",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 21 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "tomo_context_used_tokens / on(session) tomo_context_max_tokens",
+          "legendFormat": "{{session}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": { "unit": "percentunit", "min": 0, "max": 1 },
+        "overrides": []
+      }
+    },
+    {
+      "id": 14,
+      "type": "timeseries",
+      "title": "Tool calls per hour",
+      "gridPos": { "h": 8, "w": 12, "x": 0, "y": 29 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (tool) (increase(tomo_tool_calls_total[1h]))",
+          "legendFormat": "{{tool}}"
+        }
+      ],
+      "fieldConfig": {
+        "defaults": { "unit": "short", "custom": { "stacking": { "mode": "normal" } } },
+        "overrides": []
+      }
+    },
+    {
+      "id": 15,
+      "type": "timeseries",
+      "title": "Messages per hour by channel",
+      "gridPos": { "h": 8, "w": 12, "x": 12, "y": 29 },
+      "datasource": { "type": "prometheus", "uid": "prometheus" },
+      "targets": [
+        {
+          "refId": "A",
+          "expr": "sum by (channel, role) (increase(tomo_messages_total[1h]))",
+          "legendFormat": "{{channel}} / {{role}}"
+        }
+      ],
+      "fieldConfig": { "defaults": { "unit": "short" }, "overrides": [] }
+    },
+    {
+      "id": 20,
+      "type": "logs",
+      "title": "Activity feed",
+      "gridPos": { "h": 12, "w": 24, "x": 0, "y": 37 },
+      "datasource": { "type": "loki", "uid": "loki" },
+      "targets": [{ "refId": "A", "expr": "{job=\"tomo-activity\"}" }],
+      "options": {
+        "showTime": true,
+        "sortOrder": "Descending",
+        "wrapLogMessage": true,
+        "enableLogDetails": true
+      }
+    }
+  ],
+  "annotations": {
+    "list": [
+      {
+        "name": "Compactions",
+        "datasource": { "type": "loki", "uid": "loki" },
+        "enable": true,
+        "expr": "{job=\"tomo-activity\", type=\"compact\"}",
+        "iconColor": "orange"
+      },
+      {
+        "name": "Heartbeats",
+        "datasource": { "type": "loki", "uid": "loki" },
+        "enable": true,
+        "expr": "{job=\"tomo-activity\", type=\"heartbeat\"}",
+        "iconColor": "blue"
+      },
+      {
+        "name": "Issues",
+        "datasource": { "type": "loki", "uid": "loki" },
+        "enable": false,
+        "expr": "{job=\"tomo-activity\", type=\"issue\"}",
+        "iconColor": "red"
+      }
+    ]
+  }
+}

package/contrib/observability/grafana/provisioning/dashboards/dashboards.yml

@@ -0,0 +1,8 @@
+apiVersion: 1
+
+providers:
+  - name: tomo
+    folder: ""
+    type: file
+    options:
+      path: /var/lib/grafana/dashboards

package/contrib/observability/grafana/provisioning/datasources/datasources.yml

@@ -0,0 +1,16 @@
+apiVersion: 1
+
+datasources:
+  - name: Prometheus
+    uid: prometheus
+    type: prometheus
+    access: proxy
+    url: http://prometheus:9090
+    isDefault: true
+    editable: false
+  - name: Loki
+    uid: loki
+    type: loki
+    access: proxy
+    url: http://loki:3100
+    editable: false

package/contrib/observability/prometheus.yml

@@ -0,0 +1,9 @@
+global:
+  scrape_interval: 15s
+  evaluation_interval: 15s
+
+scrape_configs:
+  - job_name: tomo
+    static_configs:
+      # Tomo's metrics exporter on the Docker host (metrics.port, default 9464).
+      - targets: ["host.docker.internal:9464"]

package/dist/agent/delivery-pipeline.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"delivery-pipeline.d.ts","sourceRoot":"","sources":["../../src/agent/delivery-pipeline.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAGtE,OAAO,EAAE,kBAAkB,EAAiB,MAAM,iBAAiB,CAAC;AAGpE,UAAU,oBAAoB;IAC5B,qBAAqB,CAAC,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;CACvE;AAED,KAAK,iBAAiB,GAAG,UAAU,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE/D,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAM9D;AAED,qBAAa,gBAAgB;IACf,OAAO,CAAC,IAAI;gBAAJ,IAAI,EAAE,oBAAoB;IAE9C;;;;;;OAMG;IACG,eAAe,CACnB,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,OAAO,EACrB,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,gBAAgB,EACxB,QAAQ,GAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,OAAuB,GACtD,OAAO,CAAC,IAAI,CAAC;IAoBhB;;;OAGG;IACG,uBAAuB,CAC3B,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,MAAM,GAAE,iBAA4C,GACnD,OAAO,CAAC,IAAI,CAAC;IAoBhB;;;;OAIG;IACH,gBAAgB,CACd,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,gBAAgB,GACvB,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC;CAsBnC"}
+{"version":3,"file":"delivery-pipeline.d.ts","sourceRoot":"","sources":["../../src/agent/delivery-pipeline.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,OAAO,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAGtE,OAAO,EAA2B,kBAAkB,EAAiB,MAAM,iBAAiB,CAAC;AAG7F,UAAU,oBAAoB;IAC5B,qBAAqB,CAAC,UAAU,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;CACvE;AAED,KAAK,iBAAiB,GAAG,UAAU,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE/D,wBAAgB,oBAAoB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAM9D;AAED,qBAAa,gBAAgB;IACf,OAAO,CAAC,IAAI;gBAAJ,IAAI,EAAE,oBAAoB;IAE9C;;;;;;OAMG;IACG,eAAe,CACnB,UAAU,EAAE,MAAM,EAClB,YAAY,EAAE,OAAO,EACrB,WAAW,EAAE,MAAM,EACnB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,gBAAgB,EACxB,QAAQ,GAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,OAAuB,GACtD,OAAO,CAAC,IAAI,CAAC;IAoBhB;;;OAGG;IACG,uBAAuB,CAC3B,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,MAAM,GAAE,iBAA4C,GACnD,OAAO,CAAC,IAAI,CAAC;IAoBhB;;;;OAIG;IACH,gBAAgB,CACd,OAAO,EAAE,OAAO,EAChB,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,gBAAgB,GACvB,CAAC,IAAI,EAAE,MAAM,KAAK,OAAO,CAAC,IAAI,CAAC;CA4BnC"}