@hera-al/server 1.6.12 → 1.6.13

package/installationPkg/config.example.yaml

@@ -7,7 +7,7 @@ host: "127.0.0.1"                         # bind address for HTTP servers
 logLevel: "info"                          # debug | info | warn | error
 verboseDebugLogs: true                    # show full prompt/response text in debug logs (false = truncate to 15 chars)
 timezone: ""                              # auto-detected on first boot (e.g. Europe/Rome, America/New_York)
-fastProxyUrl: "http://localhost:4181"     # URL of the fast proxy server for proxied external models
+fastProxyUrl: "http://localhost:4181"      # Anthropic Tool Name proxy URL (atn-proxy)
 
 channels:
   telegram:
@@ -27,7 +27,7 @@ channels:
 
         # reactionLevel: "ack"       # agent reaction capability (default: "ack")
         #   off       - no reactions
-        #   ack       - bot shows 👀 while processing
+        #   ack       - bot shows eye while processing
         #   minimal   - agent can react sparingly (~1 per 5-10 messages)
         #   extensive - agent can react liberally when appropriate
 
@@ -126,25 +126,128 @@ channels:
 
   responses:
     enabled: true
-    port: 3000
+    port: 3004
 
 models:                                    # model registry
-  - id: "claude-opus-4-6"
-    name: "Claude Opus"
-    types: [internal]                      # internal | external
+  # ─── Internal models (Anthropic, via SDK) ──────────────────────
+  - id: claude-opus-4-6
+    name: Claude Opus
+    types: [internal]                      # internal | external | env-var
+  - id: claude-sonnet-4-6
+    name: Claude Sonnet
+    types: [internal]
+  - id: claude-haiku-3-5-20241022
+    name: Claude Haiku
+    types: [internal]
+
+  # ─── External models (direct API) ─────────────────────────────
   - id: openai
     name: OpenAI
-    types:
-      - external
-    proxy: "not-used"                      # not-used | direct | proxied
-    fastUrl: ""                            # OpenAI-compatible proxy URL (used when proxy is direct or proxied)
-    fastProxyApiKey: ""                    # API key for the proxy endpoint (sk-...)
-    apiKey: ""
+    types: [external]
+    apiKey: "${OPENAI_API_KEY}"
     baseURL: https://api.openai.com/v1
     useEnvVar: ""
+    contextWindow: 200000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: embeddinggemma
+    name: Embedding Gemma
+    types: [external]
+    apiKey: "${EMBEDDING_GEMMA}"
+    baseURL: http://localhost:11434/api     # local Ollama endpoint
+    useEnvVar: EMBEDDING_GEMMA
+    contextWindow: 8192
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+
+  # ─── Env-var entries (credential storage) ──────────────────────
+  - id: GOOGLE_CLIENT_ID
+    name: Google Client ID
+    types: [env-var]
+    apiKey: "${GOOGLE_CLIENT_ID}"
+    baseURL: ""
+    useEnvVar: GOOGLE_CLIENT_ID
+  - id: GOOGLE_CLIENT_SECRET
+    name: Google Client Secret
+    types: [env-var]
+    apiKey: "${GOOGLE_CLIENT_SECRET}"
+    baseURL: ""
+    useEnvVar: GOOGLE_CLIENT_SECRET
+  - id: OPENROUTER_API_KEY
+    name: OpenRouter Key
+    types: [env-var]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: ""
+    useEnvVar: OPENROUTER_API_KEY
+
+  # ─── OpenRouter models (external, via openrouter.ai) ───────────
+  - id: "openrouter:openai/gpt-5.2"
+    name: OpenRouter GPT
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 400000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: "openrouter:google/gemini-3-flash-preview"
+    name: OpenRouter Flash
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 1000000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: "openrouter:openai/gpt-4.1-mini"
+    name: OpenRouter RollingMem
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 1000000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: "openrouter:x-ai/grok-4.1-fast"
+    name: OpenRouter Grok
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 2000000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: "openrouter:google/gemini-3-pro-preview"
+    name: OpenRouter GeminiPro
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 1000000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
+  - id: "openrouter:anthropic/claude-opus-4.6"
+    name: OpenRouter Opus1m
+    types: [external]
+    apiKey: "${OPENROUTER_API_KEY}"
+    baseURL: https://openrouter.ai/api/v1
+    contextWindow: 1000000
+    costInput: 0
+    costOutput: 0
+    costCacheRead: 0
+    costCacheWrite: 0
 
 stt:
-  enabled: false
+  enabled: true
   provider: openai-whisper                 # openai-whisper | local-whisper
   openai-whisper:
     modelRef: "OpenAI"                     # references a model name in the models registry (for API key)
@@ -155,7 +258,7 @@ stt:
     model: "base"
 
 tts:
-  enabled: false
+  enabled: true
   provider: openai                         # openai | edge | elevenlabs
   maxTextLength: 4096
   timeoutMs: 30000
@@ -166,7 +269,7 @@ tts:
   openai:
     modelRef: "OpenAI"                     # references a model name in the models registry (for API key)
     model: gpt-4o-mini-tts                 # actual model ID sent to the OpenAI API
-    voice: alloy
+    voice: echo
   elevenlabs:
     modelRef: ""
     voiceId: pMsXgVXv3BLzUgSXRplE
@@ -174,31 +277,38 @@ tts:
 
 memory:
   enabled: true
-  recallStrategy: "builtin-only"           # builtin-only | search
+  recallStrategy: "search"                 # builtin-only | search
   search:
-    enabled: false                         # enable hybrid BM25 + semantic search over memory
-    modelRef: ""                           # references a model in the models registry (for API key + base URL)
+    enabled: true                          # enable hybrid BM25 + semantic search over memory
+    modelRef: "Embedding Gemma"            # references a model in the models registry (for API key + base URL)
     embeddingModel: "text-embedding-3-small"
     prefixQuery: ""                        # prefix template for query embeddings (use {content} placeholder)
     prefixDocument: ""                     # prefix template for document embeddings (use {content} placeholder)
     embeddingDimensions: 1536              # 512, 768, or 1536 (native) — lower = faster, higher = more precise
     updateDebounceMs: 3000                 # debounce for fs.watch before re-indexing
     embedIntervalMs: 300000                # interval (ms) between embedding cycles (5 min)
-    maxResults: 10                         # default max results returned by memory_search
+    maxResults: 6                          # default max results returned by memory_search
     maxSnippetChars: 700                   # max chars per snippet in results
     maxInjectedChars: 4000                 # max total chars injected into context from search results
     rrfK: 60                               # RRF fusion constant (higher = smoother blending)
-  dir: "./memory"                          # memory storage directory (relative to gmabPath/data)
 
 agent:
   model: "Claude Opus"                     # model name or ID from the models registry (format: "Name:id")
   mainFallback: ""                         # fallback model if primary fails (name or ID)
+  # engine — agent execution engine
+  engine:
+    type: "claudecode"                     # claudecode | pi
+    piModelRef: ""                         # for pi engine: "Name:provider:modelId" (e.g. "GPT:openrouter:openai/gpt-5.2")
   # picoAgent — use external LLM providers (OpenRouter, OpenAI, etc.) via pi-ai instead of Claude SDK
   picoAgent:
-    enabled: false
-    modelRefs: []                          # list of "Name:provider:modelId" refs (e.g. "GPT:openrouter:openai/gpt-5.2")
-    rollingMemoryModel: ""                 # model for context summarization, full ref format (e.g. "MyModel:openrouter:openai/gpt-4.1-mini")
-  maxTurns: 30
+    enabled: true
+    modelRefs:                             # list of "Name:provider:modelId" refs
+      - "Flash:openrouter:google/gemini-3-flash-preview"
+      - "GeminiPro:openrouter:google/gemini-3-pro-preview"
+      - "Grok:openrouter:x-ai/grok-4.1-fast"
+      - "Opus1m:openrouter:anthropic/claude-opus-4.6"
+    rollingMemoryModel: "RollingMem:openrouter:openai/gpt-4.1-mini"
+  maxTurns: 300
   permissionMode: "bypassPermissions"
   sessionTTL: 3600
   workspacePath: "./workspace"             # agent's working directory
@@ -216,16 +326,17 @@ agent:
   queueDropPolicy: "summarize"
   allowedTools:
     - "Read"
-    - "Grep"
-    - "Bash"
-    - "WebSearch"
-    - "Glob"
     - "Write"
     - "Edit"
+    - "Bash"
+    - "Glob"
+    - "Grep"
+    - "WebSearch"
     - "WebFetch"
     - "Task"
     - "Skill"
-  disallowedTools: []
+  disallowedTools:
+    - NotebookEdit
   mcpServers: {}
   builtinCoderSkill: false                 # enable coder skill in system prompt (toggle at runtime with /coder on|off)
   settingSources: project                  # SDK settings to load: nothing | user | project | both
@@ -318,7 +429,7 @@ agent:
   plugins: []                              # local plugins, configurable via Nostromo UI
   inflightTyping: true                     # keep typing indicator active when multiple messages are in-flight
   autoApproveTools: true                   # auto-approve SDK tool permissions; when false, asks user via channel buttons
-  autoRenew: 0                             # auto-renew session after N consecutive errors (0 = disabled)
+  autoRenew: 4                             # auto-renew session after N consecutive errors (0 = disabled)
 
 cron:
   enabled: true
@@ -326,7 +437,7 @@ cron:
   broadcastEvents: false                   # deliver cron responses to all known chats across all active channels
   # storePath: ""                          # leave empty for default (data/cron/jobs.json)
   heartbeat:
-    enabled: false
+    enabled: true
     every: 1800000
     channel: ""
     chatId: ""
@@ -336,10 +447,10 @@ cron:
 nostromo:
   enabled: true
   port: 3001
-  basePath: "/nostromo"                    # base URL path for the admin UI (e.g. /nostromo → http://host:port/nostromo)
+  basePath: "/nostromo"                    # base URL path for the admin UI (e.g. /nostromo -> http://host:port/nostromo)
   configCheckInterval: 5                   # seconds between config file checks
   autoRestart: true                        # auto-restart server when config.yaml changes
 
 browser:
-    enabled: false
-    controlPort: 3003
+  enabled: false
+  controlPort: 3003

package/installationPkg/default-jobs.json

@@ -0,0 +1,77 @@
+{
+  "version": 1,
+  "jobs": [
+    {
+      "id": "cron-dreaming-default",
+      "name": "dreaming",
+      "description": "Nightly memory consolidation — rereads past week's conversations, identifies patterns, updates MEMORY.md + concept map, writes dream journal",
+      "enabled": true,
+      "isolated": true,
+      "suppressToken": true,
+      "createdAtMs": 0,
+      "updatedAtMs": 0,
+      "schedule": {
+        "kind": "cron",
+        "expr": "0 3 * * *"
+      },
+      "channel": "",
+      "chatId": "",
+      "message": "It's 03:00 — time to dream. Load the dreaming skill and execute the full nightly cycle: gather the past week's conversations, digest patterns, synthesize insights, write the dream journal, update MEMORY.md and the concept map. Only message the user if you discover something genuinely interesting.",
+      "state": {}
+    },
+    {
+      "id": "cron-buongiorno-default",
+      "name": "buongiorno",
+      "description": "Morning standup: delivery report from overnight work, service health, weather, calendar, priorities for the day.",
+      "enabled": true,
+      "isolated": true,
+      "suppressToken": true,
+      "createdAtMs": 0,
+      "updatedAtMs": 0,
+      "schedule": {
+        "kind": "cron",
+        "expr": "0 7 * * *"
+      },
+      "channel": "",
+      "chatId": "",
+      "message": "It's 07:00 — time for the morning standup. Load the \"buongiorno\" skill and execute the full morning cycle. PRIORITY ORDER: 1) Check HEARTBEAT.md and workspace for overnight work results — report what was DELIVERED. 2) Quick service health check: run Google scripts LOCALLY (gmail.sh, drive.sh — they run on this server, NOT on remote nodes). Check connected nodes with list_nodes ONLY as info — if a node is offline, just note it and MOVE ON immediately. Do NOT wait, do NOT retry. 3) Calendar events and reminders for today (Google Calendar via local scripts; skip Apple Reminders if the Mac node is offline). 4) Weather for the user's city (condensed). 5) Suggest 2-3 priorities for the day. Compose a concise standup message focusing on DELIVERABLES and send it to the user. Update HEARTBEAT.md with today's task checklist. Log results in workspace/agent-thoughts/buongiorno-log.md. CRITICAL: This job has a 10-minute timeout. Do NOT spend more than 30 seconds on any single tool call. If something hangs, skip it and move on.",
+      "state": {}
+    },
+    {
+      "id": "cron-sera-default",
+      "name": "sera",
+      "description": "Evening task collector and overnight work planner. Collects open tasks, plans autonomous overnight work, updates HEARTBEAT.md for morning standup.",
+      "enabled": true,
+      "isolated": true,
+      "suppressToken": true,
+      "createdAtMs": 0,
+      "updatedAtMs": 0,
+      "schedule": {
+        "kind": "cron",
+        "expr": "0 22 * * *"
+      },
+      "channel": "",
+      "chatId": "",
+      "message": "It's 23:00 — time for the evening review. Load the \"sera\" skill and execute the full evening cycle: scan today's conversations and HEARTBEAT.md for open tasks, triage them (autonomous tonight / needs user / blocked), execute what you can autonomously, update HEARTBEAT.md with overnight work status and tomorrow's checklist. Only message the user if you have significant overnight work planned or completed something noteworthy. Log results in workspace/agent-thoughts/sera-log.md.",
+      "state": {}
+    },
+    {
+      "id": "cron-librarian-default",
+      "name": "librarian_v2",
+      "description": "Memory hygiene: deduplicate, clean, and organize MEMORY.md and concept map. Runs Wed + Sun at 04:00 (after dreaming).",
+      "enabled": true,
+      "isolated": true,
+      "suppressToken": true,
+      "createdAtMs": 0,
+      "updatedAtMs": 0,
+      "schedule": {
+        "kind": "cron",
+        "expr": "0 4 * * 0,3"
+      },
+      "channel": "",
+      "chatId": "",
+      "message": "It's 04:00 — time for memory housekeeping. Load the librarian skill and execute the full cycle:\n\n1. Read MEMORY.md completely — note the line count\n2. Read conversation logs from the last 3-4 days (memory/**/*.md, skip dreams/ and librarian/ subfolders)\n3. Read the latest dream journal from memory/dreams/\n4. Read the latest librarian report from memory/librarian/ (if any exists)\n5. **MANDATORY BACKUP**: Before ANY edit, copy MEMORY.md to memory/librarian/MEMORY.md.YYYY-MM-DD.bak (use today's date). If the copy fails, STOP and do not edit. Report the failure.\n6. Scan MEMORY.md for: duplicated information, stale TODOs (completed or abandoned >2 weeks), contradictions, outdated facts, session entries that can be compressed\n7. Review the [CONCEPTS] and [CONCEPT_MAP] RDF sections for: orphan concepts, orphan references, duplicate triples, stale relationships, naming inconsistencies\n8. Apply changes surgically to MEMORY.md — remove clear duplicates, compress old session entries, fix contradictions, clean concept map. When in doubt, DO NOT remove — flag it instead.\n9. Write a librarian report to memory/librarian/YYYY-MM-DD.md (use today's date) with: lines before/after, changes made, items flagged for review, health assessment\n10. ONLY message the user if major cleanup was done (>20 lines removed), contradictions were found, or items need their decision. Otherwise stay silent.",
+      "state": {}
+    }
+  ]
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hera-al/server",
-  "version": "1.6.12",
+  "version": "1.6.13",
   "private": false,
   "description": "Hera Artificial Life — Multi-channel AI agent gateway with autonomous capabilities",
   "license": "MIT",
@@ -62,7 +62,8 @@
     "restore-workspace": "./scripts/restore-workspace.sh"
   },
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "latest",
+    "@a2ui/lit": "^0.8.1",
+    "@anthropic-ai/claude-agent-sdk": "^0.2.50",
     "@clack/prompts": "^1.0.0",
     "@grammyjs/runner": "^2.0.3",
     "@hera-al/browser-server": "^1.0.5",