intentkit 1.2.1__tar.gz → 1.2.3__tar.gz

{intentkit-1.2.1 → intentkit-1.2.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: intentkit
-Version: 1.2.1
+Version: 1.2.3
 Summary: Intent-based AI Agent Platform - Core Package
 Project-URL: Homepage, https://github.com/crestalnetwork/intentkit
 Project-URL: Repository, https://github.com/crestalnetwork/intentkit

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/__init__.py

@@ -3,7 +3,7 @@
 A powerful platform for building AI agents with blockchain and cryptocurrency capabilities.
 """
 
-__version__ = "1.2.1"
+__version__ = "1.2.3"
 __author__ = "Ruihua"
 __email__ = "ruihua@crestal.network"
 

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/engine.py

@@ -65,11 +65,71 @@ logger = logging.getLogger(__name__)
 _FORWARDABLE_TYPES = frozenset(
     {
         ChatMessageAttachmentType.IMAGE,
+        ChatMessageAttachmentType.AUDIO,
         ChatMessageAttachmentType.VIDEO,
         ChatMessageAttachmentType.FILE,
     }
 )
 
+# Maps each forwardable attachment type to (model capability flag,
+# unsupported-error system message). Image keeps the legacy OpenAI-style
+# `image_url` block, accepted by every LangChain provider adapter we use.
+# The other three emit LangChain v0.3 standard multimodal blocks
+# `{"type": ..., "url": ...}` — currently only langchain-google-genai 3.2+
+# converts these to provider Parts. See _MEDIA_PROVIDER_SUPPORT.
+_MEDIA_INPUT_SPECS: list[tuple[ChatMessageAttachmentType, str, SystemMessageType]] = [
+    (
+        ChatMessageAttachmentType.IMAGE,
+        "supports_image_input",
+        SystemMessageType.IMAGE_INPUT_NOT_SUPPORTED,
+    ),
+    (
+        ChatMessageAttachmentType.AUDIO,
+        "supports_audio_input",
+        SystemMessageType.AUDIO_INPUT_NOT_SUPPORTED,
+    ),
+    (
+        ChatMessageAttachmentType.VIDEO,
+        "supports_video_input",
+        SystemMessageType.VIDEO_INPUT_NOT_SUPPORTED,
+    ),
+    (
+        ChatMessageAttachmentType.FILE,
+        "supports_file_input",
+        SystemMessageType.FILE_INPUT_NOT_SUPPORTED,
+    ),
+]
+
+# Providers whose LangChain adapter accepts our content-block shape for each
+# media type. Non-image types use the v0.3 standard block which currently
+# only langchain-google-genai handles; OpenAI/Anthropic require base64 data
+# blocks (`input_audio` / `document`) and are not yet implemented. A model
+# advertising capability=true on an unsupported provider is gated the same
+# way as one that doesn't advertise the capability — better than emitting
+# blocks that crash mid-stream with AGENT_INTERNAL_ERROR.
+_MEDIA_PROVIDER_SUPPORT: dict[ChatMessageAttachmentType, set[LLMProvider] | None] = {
+    ChatMessageAttachmentType.IMAGE: None,  # all providers
+    ChatMessageAttachmentType.AUDIO: {LLMProvider.GOOGLE},
+    ChatMessageAttachmentType.VIDEO: {LLMProvider.GOOGLE},
+    ChatMessageAttachmentType.FILE: {LLMProvider.GOOGLE},
+}
+
+
+def _model_can_deliver_media(
+    atype: ChatMessageAttachmentType, provider: LLMProvider
+) -> bool:
+    allowed = _MEDIA_PROVIDER_SUPPORT.get(atype)
+    return allowed is None or provider in allowed
+
+
+def _build_media_content_block(
+    atype: ChatMessageAttachmentType, url: str
+) -> dict[str, Any]:
+    if atype == ChatMessageAttachmentType.IMAGE:
+        return {"type": "image_url", "image_url": {"url": url}}
+    return {"type": atype.value, "url": url}
+
+
 # Cap raw_chunks to prevent unbounded memory growth in super_mode
 _MAX_RAW_CHUNKS = 200
 
@@ -810,24 +870,30 @@ async def stream_agent_raw(
 
     last = start
 
-    # Extract images from attachments
-    image_urls = []
+    # Group media URLs by attachment type and gate each on the model's
+    # corresponding supports_<type>_input capability.
+    media_urls_by_type: dict[ChatMessageAttachmentType, list[str]] = {
+        atype: [] for atype, _, _ in _MEDIA_INPUT_SPECS
+    }
     if user_message.attachments:
-        image_urls = [
-            str(att["url"])
-            for att in user_message.attachments
-            if "type" in att
-            and att["type"] == ChatMessageAttachmentType.IMAGE
-            and "url" in att
-            and att["url"] is not None
-        ]
-    if image_urls and not model.supports_image_input:
-        yield await _create_system_error_response(
-            SystemMessageType.IMAGE_INPUT_NOT_SUPPORTED,
-            user_message,
-            time.perf_counter() - start,
-        )
-        return
+        for att in user_message.attachments:
+            atype = att.get("type")
+            url = att.get("url")
+            if atype in media_urls_by_type and url is not None:
+                media_urls_by_type[atype].append(str(url))
+
+    for atype, capability, error_type in _MEDIA_INPUT_SPECS:
+        if not media_urls_by_type[atype]:
+            continue
+        if not getattr(model, capability) or not _model_can_deliver_media(
+            atype, model.provider
+        ):
+            yield await _create_system_error_response(
+                error_type,
+                user_message,
+                time.perf_counter() - start,
+            )
+            return
 
     input_message = user_message.message
 
@@ -861,20 +927,22 @@ async def stream_agent_raw(
     messages = [
         HumanMessage(content=input_message),
     ]
-    if image_urls:
+    for atype, _, _ in _MEDIA_INPUT_SPECS:
+        urls = media_urls_by_type[atype]
+        if not urls:
+            continue
         logger.info(
-            "Passing %d image url(s) to LLM for agent=%s chat=%s: %s",
-            len(image_urls),
+            "Passing %d %s url(s) to LLM for agent=%s chat=%s: %s",
+            len(urls),
+            atype.value,
             user_message.agent_id,
             user_message.chat_id,
-            image_urls,
+            urls,
         )
         messages.extend(
             [
-                HumanMessage(
-                    content=[{"type": "image_url", "image_url": {"url": image_url}}]
-                )
-                for image_url in image_urls
+                HumanMessage(content=[_build_media_content_block(atype, url)])
+                for url in urls
             ]
         )
 

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/lead/engine.py

@@ -70,8 +70,8 @@ async def _build_lead_agent(team_id: str) -> Agent:
     prompt = (
         "### Sub-Agents\n\n"
         "Use `lead_call_agent` to delegate:\n\n"
-        "- `agent-manager`: Agent CRUD.\n"
-        "- `task-manager`: Autonomous task scheduling and management.\n"
+        "- `agent-manager`: Manage team agents — create, configure, update, "
+        "and schedule autonomous (cron) tasks on them.\n"
         "- `self-updater`: Update your own name, avatar, personality, or memory.\n"
         "- `content-manager`: Read team activities and posts.\n"
         "- `user-manager`: Update the current user's profile (name, timezone, language).\n\n"

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/lead/skills/call_agent.py

@@ -35,7 +35,7 @@ class LeadCallAgentInput(BaseModel):
     message: str = Field(..., description="Message to send")
     attachments: list[AttachmentRef] | None = Field(
         None,
-        description="Optional attachments (images, videos, files) to forward to the sub-agent. Use when delegating tasks that need media from previous messages.",
+        description="Optional attachments (images, audio, videos, files) to forward to the sub-agent. Use when delegating tasks that need media from previous messages.",
     )
 
 
@@ -43,7 +43,7 @@ class LeadCallAgent(LeadSkill):
     """Team-aware call_agent that supports both in-memory sub-agents and DB agents.
 
     Resolution order:
-    1. Check in-memory sub-agent registry (agent-manager, task-manager, etc.)
+    1. Check in-memory sub-agent registry (agent-manager, self-updater, etc.)
     2. Fall back to database agent lookup (scoped to team)
     """
 

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/lead/sub_agents/__init__.py

@@ -20,7 +20,6 @@ logger = logging.getLogger(__name__)
 
 # Sub-agent slug constants
 SLUG_AGENT_MANAGER = "agent-manager"
-SLUG_TASK_MANAGER = "task-manager"
 SLUG_SELF_UPDATER = "self-updater"
 SLUG_CONTENT_MANAGER = "content-manager"
 SLUG_USER_MANAGER = "user-manager"
@@ -109,10 +108,6 @@ from intentkit.core.lead.sub_agents.self_updater import (  # noqa: E402
     build_self_updater,
     get_self_updater_skills,
 )
-from intentkit.core.lead.sub_agents.task_manager import (  # noqa: E402
-    build_task_manager,
-    get_task_manager_skills,
-)
 from intentkit.core.lead.sub_agents.user_manager import (  # noqa: E402
     build_user_manager,
     get_user_manager_skills,
@@ -122,20 +117,13 @@ SUB_AGENT_REGISTRY: dict[str, SubAgentDefinition] = {
     SLUG_AGENT_MANAGER: SubAgentDefinition(
         slug=SLUG_AGENT_MANAGER,
         description=(
-            "Manages team agents: create, update, configure, list agents. "
-            "Also provides LLM model info and available skills for agent configuration."
+            "Manages team agents end-to-end: create, update, configure, list "
+            "agents, and schedule autonomous tasks on them. Also exposes LLM "
+            "model info and available skills for agent configuration."
         ),
         build_fn=build_agent_manager,
         skills_fn=get_agent_manager_skills,
     ),
-    SLUG_TASK_MANAGER: SubAgentDefinition(
-        slug=SLUG_TASK_MANAGER,
-        description=(
-            "Manages autonomous tasks: list, add, edit, delete scheduled tasks for agents."
-        ),
-        build_fn=build_task_manager,
-        skills_fn=get_task_manager_skills,
-    ),
     SLUG_SELF_UPDATER: SubAgentDefinition(
         slug=SLUG_SELF_UPDATER,
         description=(

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/lead/sub_agents/agent_manager.py

@@ -1,4 +1,4 @@
-"""Agent Manager sub-agent definition."""
+"""Agent Manager sub-agent: agent CRUD plus autonomous tasks scheduled on those agents."""
 
 from __future__ import annotations
 
@@ -7,9 +7,21 @@ from datetime import datetime, timezone
 
 from langchain_core.tools import BaseTool
 
+from intentkit.core.lead.skills.add_autonomous_task import (
+    lead_add_autonomous_task_skill,
+)
 from intentkit.core.lead.skills.create_team_agent import create_team_agent_skill
+from intentkit.core.lead.skills.delete_autonomous_task import (
+    lead_delete_autonomous_task_skill,
+)
+from intentkit.core.lead.skills.edit_autonomous_task import (
+    lead_edit_autonomous_task_skill,
+)
 from intentkit.core.lead.skills.get_team_agent import get_team_agent_skill
 from intentkit.core.lead.skills.get_team_info import get_team_info_skill
+from intentkit.core.lead.skills.list_autonomous_tasks import (
+    lead_list_autonomous_tasks_skill,
+)
 from intentkit.core.lead.skills.list_skills import lead_list_available_skills_skill
 from intentkit.core.lead.skills.list_team_agents import list_team_agents_skill
 from intentkit.core.lead.skills.llm import lead_get_available_llms_skill
@@ -28,6 +40,10 @@ def get_agent_manager_skills() -> Sequence[BaseTool]:
         update_team_agent_skill,
         lead_get_available_llms_skill,
         lead_list_available_skills_skill,
+        lead_list_autonomous_tasks_skill,
+        lead_add_autonomous_task_skill,
+        lead_edit_autonomous_task_skill,
+        lead_delete_autonomous_task_skill,
     ]
 
 
@@ -76,6 +92,37 @@ def build_agent_manager(team_id: str) -> Agent:
         "- Access levels: `public` (all users), `private` (owner only)\n"
         "- Only include skills you want to enable — omitted skills stay disabled\n"
         "- The backend will reject unknown categories or skill names with an error\n\n"
+        "### Internet Search\n\n"
+        "To give an agent web search ability, set the agent field "
+        "`search_internet` to `true`. That switch enables the LLM provider's "
+        "native web search and is the correct way to add general-purpose "
+        "search. Do NOT add categories like `tavily`, `firecrawl`, "
+        "`web_scraper`, etc. just to grant search — those are backups for "
+        "specialised scraping/extraction needs and only belong in `skills` "
+        "when the agent really needs them.\n\n"
+        "### Autonomous Tasks\n\n"
+        "Tasks belong to an agent and run on a cron schedule. Use these tools:\n"
+        "- `lead_list_autonomous_tasks` — list tasks on an agent (call before "
+        "edit/delete to discover task IDs).\n"
+        "- `lead_add_autonomous_task` — schedule a new task.\n"
+        "- `lead_edit_autonomous_task` — update fields on an existing task.\n"
+        "- `lead_delete_autonomous_task` — remove a task.\n\n"
+        "Workflow:\n"
+        "1. If the user did not name an agent, call `lead_list_team_agents`.\n"
+        "2. Call `lead_list_autonomous_tasks` to see existing tasks before "
+        "adding/editing/deleting.\n"
+        "3. Apply the change.\n\n"
+        "Cron expressions (5 fields: min hour day month weekday):\n"
+        "- `*/5 * * * *` — every 5 min\n"
+        "- `0 */2 * * *` — every 2 hours\n"
+        "- `0 9 * * *` — daily 9:00 UTC\n"
+        "- `0 9 * * 1-5` — weekdays 9:00 UTC\n\n"
+        "Conditional tasks: for condition-based work, schedule a polling "
+        "task (e.g. every 5 min). Unless the user wants it to run forever, "
+        "tell the task to delete itself once the condition is met.\n\n"
+        "Tips:\n"
+        "- Set `has_memory=True` only when the task needs context between runs.\n"
+        "- Prefer disabling (`enabled=False`) over deleting for temporary pauses.\n\n"
         "### Agent Fields Reference\n\n"
         "- `name`: Display name (max 50 chars)\n"
         "- `purpose`, `personality`, `principles`: Agent character\n"
@@ -99,9 +146,12 @@ def build_agent_manager(team_id: str) -> Agent:
         "owner": "system",
         "team_id": team_id,
         "name": "Agent Manager",
-        "purpose": "Create, configure, and update team agents.",
+        "purpose": (
+            "Create, configure, and update team agents, including the "
+            "autonomous tasks scheduled on those agents."
+        ),
         "principles": (
-            "1. Speak to users in their language, but use English in agent configuration.\n"
+            "1. Speak to users in their language, but use English in agent and task configuration.\n"
             "2. All changes are directly deployed (no draft flow).\n"
             "3. Update is override — provide complete field values, not just changes."
         ),

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/manager/engine.py

@@ -102,6 +102,12 @@ def _build_manager_agent(agent_id: str, user_id: str) -> Agent:
         ```
         The `enabled` flag is at the category level, and `states` refers to specific skills.
         For content involving sensitive material, select `private`. For content suitable for all users, select `public`.
+
+        Internet search: to give an agent web search, set the agent field `search_internet` to `true`.
+        That switch enables the LLM provider's native web search and is the right tool for general-purpose
+        searching. Do NOT add categories like `tavily`, `firecrawl`, `web_scraper`, etc. just to give the
+        agent search ability — those categories are backups for specialised scraping/extraction needs and
+        only belong in the skills config when the agent really requires them.
         """
         "\n\n### Public Information\n\n"
         "Only agents that have already been deployed at least once can have their public"

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/manager/service.py

@@ -13,6 +13,12 @@ from fastapi import status
 
 from intentkit.core.agent import get_agent
 from intentkit.models.agent import AgentPublicInfo, AgentUserInput
+from intentkit.skills.availability import (
+    filter_unavailable_states,
+    import_skill_category,
+    is_individual_skill_available,
+    is_skill_category_available,
+)
 from intentkit.utils.error import IntentKitAPIError
 
 logger = logging.getLogger(__name__)
@@ -50,8 +56,18 @@ def agent_draft_json_schema() -> dict[str, object]:
                 if not schema_path.is_file():
                     continue
 
+                category = entry.name
+
+                module = import_skill_category(category)
+                if module is None or not is_skill_category_available(module):
+                    logger.info(
+                        "Skipped skill category '%s' for manager schema: not available",
+                        category,
+                    )
+                    continue
+
                 try:
-                    skills_properties[entry.name] = _load_skill_schema(schema_path)
+                    skill_schema = _load_skill_schema(schema_path)
                 except (
                     OSError,
                     ValueError,
@@ -59,9 +75,19 @@ def agent_draft_json_schema() -> dict[str, object]:
                     jsonref.JsonRefError,
                 ) as exc:
                     logger.warning(
-                        "Failed to load schema for skill '%s': %s", entry.name, exc
+                        "Failed to load schema for skill '%s': %s", category, exc
                     )
                     continue
+
+                schema_props = skill_schema.get("properties")
+                if isinstance(schema_props, dict):
+                    states = schema_props.get("states")
+                    if isinstance(states, dict):
+                        schema_props["states"] = filter_unavailable_states(
+                            module, category, states
+                        )
+
+                skills_properties[category] = skill_schema
     except (AttributeError, ModuleNotFoundError, ImportError):
         logger.warning("intentkit skills package not found when building schema")
         return schema
@@ -88,28 +114,38 @@ def get_skills_hierarchical_text() -> str:
                 if not schema_path.is_file():
                     continue
 
+                category = entry.name
+
+                module = import_skill_category(category)
+                if module is None or not is_skill_category_available(module):
+                    logger.info(
+                        "Skipped skill category '%s' for hierarchical text: "
+                        "not available",
+                        category,
+                    )
+                    continue
+
                 try:
                     skill_schema = _load_skill_schema(schema_path)
-                    skill_name = entry.name
                     skill_title = skill_schema.get(
-                        "title", skill_name.replace("_", " ").title()
+                        "title", category.replace("_", " ").title()
                     )
                     skill_description = skill_schema.get(
                         "description", "No description available"
                     )
                     skill_tags = cast(list[str], skill_schema.get("x-tags", ["Other"]))
 
-                    # Use the first tag as the primary category
+                    # Use the first tag as the primary group
                     primary_category = skill_tags[0] if skill_tags else "Other"
 
-                    if primary_category not in categories:
-                        categories[primary_category] = []
-
-                    # Extract individual skills from states.properties
                     individual_skills: list[dict[str, str]] = []
                     states_props = _get_states_properties(skill_schema)
                     if states_props:
                         for ind_name, ind_def in states_props.items():
+                            if not is_individual_skill_available(
+                                module, category, ind_name
+                            ):
+                                continue
                             ind_desc = (
                                 ind_def.get("description", "No description available")
                                 if isinstance(ind_def, dict)
@@ -119,9 +155,17 @@ def get_skills_hierarchical_text() -> str:
                                 {"name": ind_name, "description": ind_desc}
                             )
 
+                    # Drop the category entirely if every skill was filtered;
+                    # surfacing an empty group would just be noise to the LLM.
+                    if states_props and not individual_skills:
+                        continue
+
+                    if primary_category not in categories:
+                        categories[primary_category] = []
+
                     categories[primary_category].append(
                         {
-                            "name": skill_name,
+                            "name": category,
                             "title": skill_title,
                             "description": skill_description,
                             "individual_skills": individual_skills,
@@ -135,7 +179,7 @@ def get_skills_hierarchical_text() -> str:
                     jsonref.JsonRefError,
                 ) as exc:
                     logger.warning(
-                        "Failed to load schema for skill '%s': %s", entry.name, exc
+                        "Failed to load schema for skill '%s': %s", category, exc
                     )
                     continue
     except (AttributeError, ModuleNotFoundError, ImportError):

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/core/system_skills/call_agent.py

@@ -85,6 +85,7 @@ def render_attachments_awareness(
         match att_type:
             case (
                 ChatMessageAttachmentType.IMAGE
+                | ChatMessageAttachmentType.AUDIO
                 | ChatMessageAttachmentType.VIDEO
                 | ChatMessageAttachmentType.FILE
                 | ChatMessageAttachmentType.LINK
@@ -129,10 +130,11 @@ async def get_start_message_attachments(
     return context.start_message_attachments
 
 
-ForwardableAttachmentType = Literal["image", "video", "file"]
+ForwardableAttachmentType = Literal["image", "audio", "video", "file"]
 
 _ATTACHMENT_TYPE_MAP: dict[str, ChatMessageAttachmentType] = {
     "image": ChatMessageAttachmentType.IMAGE,
+    "audio": ChatMessageAttachmentType.AUDIO,
     "video": ChatMessageAttachmentType.VIDEO,
     "file": ChatMessageAttachmentType.FILE,
 }
@@ -142,7 +144,7 @@ class AttachmentRef(BaseModel):
     """Reference to an attachment for forwarding to another agent."""
 
     type: ForwardableAttachmentType = Field(
-        ..., description="Attachment type: image, video, or file"
+        ..., description="Attachment type: image, audio, video, or file"
     )
     url: str = Field(..., description="URL of the attachment")
 
@@ -169,7 +171,7 @@ class CallAgentInput(BaseModel):
     message: str = Field(..., description="Message to send")
     attachments: list[AttachmentRef] | None = Field(
         None,
-        description="Optional attachments (images, videos, files) to forward to the target agent. Use when delegating tasks that need media from previous messages.",
+        description="Optional attachments (images, audio, videos, files) to forward to the target agent. Use when delegating tasks that need media from previous messages.",
     )
 
 

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/models/app_setting.py

@@ -26,6 +26,9 @@ class SystemMessageType(str, Enum):
     RECURSION_LIMIT_EXCEEDED = "recursion_limit_exceeded"
     TIMEOUT_ERROR = "timeout_error"
     IMAGE_INPUT_NOT_SUPPORTED = "image_input_not_supported"
+    AUDIO_INPUT_NOT_SUPPORTED = "audio_input_not_supported"
+    VIDEO_INPUT_NOT_SUPPORTED = "video_input_not_supported"
+    FILE_INPUT_NOT_SUPPORTED = "file_input_not_supported"
 
 
 # Default system messages
@@ -40,6 +43,9 @@ DEFAULT_SYSTEM_MESSAGES = {
     "recursion_limit_exceeded": "The agent reached the maximum recursion limit. Please type 'continue' to resume execution.",
     "timeout_error": "Network request timed out or the LLM is overloaded. Please try again.",
     "image_input_not_supported": "This agent's current model does not support image input. Please switch to an image-capable model or send text instead.",
+    "audio_input_not_supported": "This agent's current model does not support audio input. Please switch to an audio-capable model or send text instead.",
+    "video_input_not_supported": "This agent's current model does not support video input. Please switch to a video-capable model or send text instead.",
+    "file_input_not_supported": "This agent's current model does not support file input. Please switch to a file-capable model or send text instead.",
 }
 
 # In-memory cache for app settings

{intentkit-1.2.1 → intentkit-1.2.3}/intentkit/models/chat.py

@@ -33,6 +33,7 @@ class ChatMessageAttachmentType(str, Enum):
 
     LINK = "link"
     IMAGE = "image"
+    AUDIO = "audio"
     VIDEO = "video"
     FILE = "file"
     XMTP = "xmtp"
@@ -65,14 +66,15 @@ class AuthorType(str, Enum):
 class ChatMessageAttachment(TypedDict):
     """Chat message attachment model.
 
-    An attachment can be a link, image, or file that is associated with a chat message.
+    An attachment can be a link, image, audio, video, or file (plus channel-
+    specific types like xmtp/card/choice) associated with a chat message.
     """
 
     type: Annotated[
         ChatMessageAttachmentType,
         Field(
             ...,
-            description="Type of the attachment (link, image, or file)",
+            description="Type of the attachment (link, image, audio, video, file, etc.)",
             examples=["link"],
         ),
     ]