-# Dana: The Worldβs First Agentic OS
+# Dana: The Cognitive Enterprise Platform
-## Build deterministic expert agent easily with Dana.
-
+> *"We have 50 years of expertise walking around in people's heads.
+> It's never been written down. It can't be searched. And every day, a little more of it disappears."*
+> β VP of Operations, Fortune 500 Manufacturer
-### A complete Expert Agent Development Toolkit: Agentic out of the box. Grounded in domain expertise.
+**What if you could capture, retain, and multiply that knowledge?**
---
-## Why Dana?
+## The $3.1 Trillion Problem
-Most frameworks make you choose:
-- **Too rigid** β narrow, specialized agents.
-- **Too generic** β LLM wrappers that fail in production.
-- **Too much glue** β orchestration code everywhere.
+Every year, enterprises lose **$3.1 trillion** to knowledge that was never captured, expertise that isn't retained, and wisdom that can't scale.
-Dana gives you the missing foundation:
+- **Knowledge never captured** β Your best operators make split-second decisions based on decades of pattern recognition. None of it is written down.
+- **Knowledge not retained** β Even when documented, context fades. The *why* behind decisions gets lost. Procedures exist but understanding doesn't.
+- **Knowledge not multiplied** β One expert can only be in one place. Their judgment doesn't scale. New hires take years to develop the same instincts.
+- **Knowledge walking out the door** β When veterans leave, retire, or move on, their expertise leaves with them.
-- **Deterministic** β flexible on input, consistent on output β reliable results every run.
-- **Contextual** β built-in memory and knowledge grounding let agents recall, adapt, and reason with domain expertise.
-- **Concurrent by default** β non-blocking execution; agents run tasks in parallel without threads or async code.
-- **Composable workflows** β chain simple steps into complex, reproducible processes that capture expert know-how.
-- **Local** β runs on your laptop or secure environments, ensuring privacy, speed, and mission-critical deployment.
-- **Robust** β fault-tolerant by design, agents recover gracefully from errors and edge cases.
-- **Adaptive** β agents learn from feedback and evolving conditions, improving performance over time.
-
+Traditional solutions don't work:
+- **Documentation?** Captures the *what*, loses the *why*. Outdated the moment it's written.
+- **Knowledge bases?** Graveyards of stale wikis nobody searches.
+- **Knowledge graphs?** Promising, but prohibitively expensive to build and maintain.
+
+**The brutal truth:** In most enterprises, critical operating knowledge exists in exactly one placeβpeople's heads. It was never captured. It's not being retained. And it certainly isn't multiplying.
---
-## Install and Launch Dana
+## What If Knowledge Could Compound?
-π‘ **Tip:** Always activate your virtual environment before running or installing anything for Dana.
+Imagine an enterprise where:
-```bash
-# Activate your virtual environment (recommended)
-source venv/bin/activate # On macOS/Linux
-# or
-venv\Scripts\activate # On Windows
+- A new engineer asks *"Why do we heat-treat at 450Β°F instead of 500Β°F?"* and gets the actual reasoningβtraced back to the 2019 incident that taught everyone that lesson.
+
+- Your AI assistant doesn't just search documentsβit *understands* how your processes connect, why decisions were made, and what happens downstream when something changes.
+
+- When regulations shift, you know instantly which procedures are affected, who owns them, and what needs to change.
+
+- Domain expertise isn't locked in veterans' headsβit's encoded, evolving, and available to every agent and every employee, 24/7.
+
+**This is the Cognitive Enterprise.** And Dana makes it possible.
+
+---
+
+## How It Works: Cognitive Ontology
+
+The secret is a new architectural layer: **Cognitive Ontology**βa living knowledge graph that captures not just *what* your enterprise knows, but *how* things connect and *why* decisions get made.
-pip install dana
-dana studio # Launch Dana Agent Studio
-dana repl # Launch Dana Repl
```
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+β TODAY: KNOWLEDGE TRAPPED β
+β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β HUMAN OPERATORS β β
+β β (context lives only in their heads) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β β
+β βΌ β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β DATA LAYER β β
+β β (databases, documents, logs β disconnected) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+
+
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+β TOMORROW: KNOWLEDGE LIBERATED β
+β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β HUMAN OPERATORS β β
+β β (amplified by encoded expertise) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β β
+β βΌ β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β COSTAR AGENTS β β
+β β (continuously build and apply knowledge) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β β
+β βΌ β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β COGNITIVE ONTOLOGY β β
+β β (living knowledge graph β built by agents, for agents) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β β
+β βΌ β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β DATA LAYER β β
+β β (now connected, contextualized, alive) β β
+β ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ β
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+```
+
+**The key insight:** Traditional knowledge graphs failed because humans had to build and maintain them. That's expensive and unsustainable.
-- For detailed setup (Python versions, OS quirks, IDE integration), see [Tech Setup](https://github.com/aitomatic/dana/blob/release/docs/tech-setup.md).
+**Dana's breakthrough:** Intelligent agents build the ontology *automatically*βextracting knowledge from documents, learning from experts, and evolving the graph continuously. The ontology is cognitive because it's created by cognition, for cognition.
---
-## Whatβs Included in v0.5
+## COSTAR: Agents That Learn
-### Agent Studio
-Turn a problem statement into a draft expert agent with three parts β agent, resources, workflows. Studio generates a best-match workflow and lets you extend it with resources (documents, generated knowledge, web search) or edit workflows directly.
+Dana agents follow the **COSTAR** lifecycleβa continuous loop of knowledge building and application:
-### Agent-Native Programming Language
-A Python-like `.na` language with a built-in runtime that provides agentic behaviors out of the box β concurrency, knowledge grounding, and deterministic execution β so you donβt have to wire these up yourself.
+```
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+β β
+β COSTAR AGENT LIFECYCLE β
+β β
+β KNOWLEDGE AGENTS COGNITIVE TASK AGENTS β
+β (build the ontology) ONTOLOGY (use the ontology) β
+β β
+β ββββββββββββ βββββββββββββββββ ββββββββββββ β
+β β CURATE ββββββββββββΆβ βββββββββββΆβ SEE β β
+β ββββββββββββ extract β Domain β context ββββββ¬ββββββ β
+β β knowledge β Knowledge β β β
+β βΌ β Graph β βΌ β
+β ββββββββββββ β β ββββββββββββ β
+β β ORGANIZE ββββββββββββΆβ βββββββββββ β β THINK β β
+β ββββββββββββ structure β β Entity β β ββββββ¬ββββββ β
+β β βββββββββββ€ β β β
+β β β Entity β β βΌ β
+β ββββββββββββ β βββββββββββ€ β ββββββββββββ β
+β β REFLECT βββββββββββββ β Entity β ββββββββββββ ACT β β
+β ββββββββββββ learning β βββββββββββ β results ββββββ¬ββββββ β
+β β² β Causal Links β β β
+β β βββββββββββββββββ βΌ β
+β β β² ββββββββββββ β
+β β βββββββββββββββββββββ REFLECT β β
+β β feedback ββββββ¬ββββββ β
+β βββββββββββββββββββββββββββββββββββββββββββββββββββ β
+β β
+ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
+```
-What this means for you: You can build and iterate on expert agents faster, with less setup and more confidence theyβll run reliably in production.
+| Phase | What Happens |
+|-------|--------------|
+| **Curate** | Agents extract knowledge from documents, interviews, and operational data |
+| **Organize** | Structure knowledge into causal and contextual relationships |
+| **See** | Perceive new situations through the lens of accumulated expertise |
+| **Think** | Reason using domain knowledge, not just pattern matching |
+| **Act** | Execute with the confidence of encoded institutional wisdom |
+| **Reflect** | Learn from outcomes, continuously improving the ontology |
-Full release notes β [v0.5 Release](https://github.com/aitomatic/dana/blob/release/docs/releases/v0.5.md).
+**The result:** Agents that don't just follow instructionsβthey *understand* your domain.
---
-## First Expert Agent in 4 Steps
+## Real-World Impact
-1. **Define an Agent**
- ```dana
- agent RiskAdvisor
- ```
+### Semiconductor Manufacturing
+*"We reduced root-cause analysis time from 3 days to 20 minutes. The system connects equipment sensor data to process outcomes in ways that took our engineers years to learn."*
-2. **Add Resources**
- ```dana
- resource_financial_docs = get_resources("rag", sources=["10-K.pdf", "Q2.xlsx"])
- ```
+### Financial Services
+*"New analysts now have access to the same contextual knowledge as our 20-year veterans. Onboarding time dropped from 6 months to 6 weeks."*
-3. **Follow an Expert Workflow**
- ```dana
- def analyze(...): return ...
- def score(...): return ...
- def recommend(...): return ...
-
- def wf_risk_check(resources) = analyze | score | recommend
+### Industrial Operations
+*"When our control system flagged an anomaly, Dana didn't just alert usβit explained why it mattered, what happened last time, and what to check first."*
- result = RiskAdvisor.solve("Identify liquidity risks", resources=[resource_financial_docs], workflows=[wf_risk_check])
-
- print(result)
- ```
+---
-4. **Run or Deploy**
- ```bash
- dana run my_agent.na # Run locally
- dana deploy my_agent.na # Deploy as REST API
- ```
+## Get Started in 5 Minutes
-
+```bash
+pip install dana
+dana studio
+```
+
+```python
+from adana.core.agent import STARAgent
+
+# Create an agent grounded in your domain knowledge
+agent = STARAgent(agent_type="operations_expert")
+
+# Point it at your knowledge sources
+agent.with_resources(
+ rag_resource("./procedures"),
+ rag_resource("./incident_reports"),
+ rag_resource("./equipment_manuals")
+)
+
+# Ask it anythingβit understands context
+result = agent.query(
+ message="Why do we use nitrogen purge before heat treatment?"
+)
+
+# Get answers with reasoning, not just retrieval
+print(result)
+# β "Nitrogen purge prevents oxide formation on titanium alloys.
+# This was established after the 2019 Q3 batch rejection (IR-2019-0847)
+# where oxide contamination caused 12% yield loss. The 15-minute purge
+# duration was determined by Process Engineering based on chamber volume
+# and acceptable O2 levels (<50ppm). See SOP-HT-003 Section 4.2."
+```
+
+---
+
+## The Inevitable Future
+
+Every enterprise will become a Cognitive Enterprise. The only question is whenβand whether you'll lead or follow.
+
+The companies building cognitive ontologies today will:
+- **Capture** expertise that was never written downβextracted by agents from experts and operations
+- **Retain** institutional knowledge that compounds over time, not fades
+- **Multiply** expert judgment across the entire organization, 24/7
+- **Evolve** as knowledge adapts with the business, not against it
+
+**Dana makes this accessible now.** Not in some distant future. Not requiring massive infrastructure investments. Today.
---
-## Learn More
+## Architecture
-- [Core Concepts](https://github.com/aitomatic/dana/blob/release/docs/core-concepts.md) β Agents, Resources, Workflows, Studio.
-- [Reference](https://github.com/aitomatic/dana/blob/release/docs/reference/language.md) β Language syntax and semantics.
-- [Primers](https://github.com/aitomatic/dana/tree/release/docs/primers) β Deep dives into Dana language design.
+```
+dana/
+βββ dana_lang/ # Language runtime & COSTAR frameworks
+βββ dana_agent/ # COSTAR agent implementation
+βββ dana_studio/ # Visual agent builder
+βββ dana/ # Contrib modules
+βββ examples/ # Ready-to-run examples
+βββ tests/ # Test suites
+βββ docs/ # Documentation
+βββ bin/ # CLI tools & scripts
+```
---
-## Community
-- π [Issues](https://github.com/aitomatic/dana/issues)
-- π¬ [Discuss on Discord](https://discord.gg/dana)
+## Learn More
+
+- [Quick Start Guide](docs/quickstart.md) β Running in 5 minutes
+- [Core Concepts](docs/core-concepts.md) β Understanding COSTAR and Cognitive Ontology
+- [Enterprise Deployment](docs/enterprise.md) β Scaling to production
+
+## Community
+
+- [GitHub Issues](https://github.com/aitomatic/dana/issues) β Report bugs, request features
+- [Discord](https://discord.gg/dana) β Join the community
+
+## Enterprise
-## Enterprise support
-- [Contact Aitomatic Sales](mailto:sales@aitomatic.com)
+Building something mission-critical? [Talk to us](mailto:sales@aitomatic.com).
---
-## License
+{user_message}
") - response_parts.append("") # Empty line for spacing - - # Add the main question - response_parts.append(f"{question}
") - response_parts.append("") # Empty line for spacing - - # Add options if provided - if options and len(options) > 0: - response_parts.append("") - response_parts.append("Or, just type your own request in the chat
") - response_parts.append("") # Empty line for spacing - # Join all parts with proper spacing - return "\n".join(response_parts) diff --git a/dana/api/services/intent_detection/intent_handlers/handler_tools/knowledge_ops_tools/attempt_completion_tool.py b/dana/api/services/intent_detection/intent_handlers/handler_tools/knowledge_ops_tools/attempt_completion_tool.py deleted file mode 100644 index 25c5f678c..000000000 --- a/dana/api/services/intent_detection/intent_handlers/handler_tools/knowledge_ops_tools/attempt_completion_tool.py +++ /dev/null @@ -1,67 +0,0 @@ -from dana.api.services.intent_detection.intent_handlers.handler_tools.base_tool import ( - BaseArgument, - BaseTool, - BaseToolInformation, - InputSchema, - ToolResult, -) - - -class AttemptCompletionTool(BaseTool): - def __init__(self): - tool_info = BaseToolInformation( - name="attempt_completion", - description="Present information to the user. Use for: 1) Final results after workflow completion, 2) Direct answers to agent information requests ('Tell me about Sofia'), 3) System capability questions ('What can you help me with?'), 4) Out-of-scope request redirection. DO NOT use for knowledge structure questions - use explore_knowledge instead. Optionally provide one option for next step if it is relevant, but if there is option provided, ALWAYS use options parameter and ONLY provided one option.", - input_schema=InputSchema( - type="object", - properties=[ - BaseArgument( - name="summary", - type="string", - description="Summary of what was accomplished, highlight the key points using bold markdown (e.g. **key points**). OR direct answer/explanation to user's question", - example="β Successfully generated 10 knowledge artifacts OR Sofia is your Personal Finance Advisor that I'm helping you build OR I specialize in building knowledge for Sofia through structure design and content generation", - ), - BaseArgument( - name="options", - type="list", - description="Provide option if there is one relevant next step or choice. Provide only ONE option. Use when presenting option to the user after completing a task or when asking for next action. Option must be a complete user response that makes sense when sent as the next message. If the summary is about added topics successfully, the option must be Generate knowledge for added topics", - example='["Add this structure to domain knowledge"]', - ), - ], - required=["summary"], - ), - ) - super().__init__(tool_info) - - def _build_interactive_response(self, summary: str, options: list[str]) -> str: - """ - Build an interactive response with HTML button-style options. - """ - response_parts = [] - - # Add the summary content - response_parts.append(f"{summary}
") - response_parts.append("") # Empty line for spacing - - # Add clickable options - response_parts.append("") - response_parts.append("Or, just type your own request in the chat
") - response_parts.append("") # Empty line for spacing - - # Join all parts with proper spacing - return "\n".join(response_parts) - - async def _execute(self, summary: str, options: list[str] = None) -> ToolResult: - """ - Execute completion with optional interactive options. - """ - if options and len(options) > 0: - content = self._build_interactive_response(summary, options) - else: - content = summary - - return ToolResult(name="attempt_completion", result=content, require_user=True) diff --git a/dana/api/services/intent_detection_service.py b/dana/api/services/intent_detection_service.py deleted file mode 100644 index 510d0edd8..000000000 --- a/dana/api/services/intent_detection_service.py +++ /dev/null @@ -1,384 +0,0 @@ -"""LLM-based Intent Detection Service for domain knowledge management.""" - -import json -import logging -from typing import Any - -import yaml -from dana.api.core.schemas import IntentDetectionRequest, IntentDetectionResponse, DomainKnowledgeTree, MessageData -from dana.common.mixins.loggable import Loggable -from dana.common.sys_resource.llm.legacy_llm_resource import LegacyLLMResource as LLMResource -from dana.common.types import BaseRequest - -logger = logging.getLogger(__name__) - - -class IntentDetectionService(Loggable): - """Service for detecting user intent in chat messages using LLM.""" - - def __init__(self): - super().__init__() - self.llm = LLMResource() - - async def detect_intent(self, request: IntentDetectionRequest) -> IntentDetectionResponse: - """Detect user intent using LLM analysis - now supports multiple intents.""" - try: - # Build the LLM prompt - prompt = self._build_intent_detection_prompt(request.user_message, request.chat_history, request.current_domain_tree) - - # Create LLM request - llm_request = BaseRequest( - arguments={ - "messages": [ - {"role": "system", "content": "You are an expert at understanding user intent in agent conversations."}, - {"role": "user", "content": prompt}, - ], - "temperature": 0.1, # Lower temperature for more consistent intent detection - "max_tokens": 500, - } - ) - - # Call LLM - response = await self.llm.query(llm_request) - - # Parse the response - try: - content = response.content - if isinstance(content, str): - result = json.loads(content) - elif isinstance(content, dict): - result = content - else: - raise ValueError(f"Unexpected LLM response type: {type(content)}") - - intent_result: dict = json.loads(result.get("choices")[0].get("message").get("content")) - - # Handle multiple intents - return the first one for backward compatibility - # but store all intents in the response - intents = intent_result.get("intents", []) - if not intents: - # Fallback to single intent format - intents = [ - { - "intent": intent_result.get("intent", "general_query"), - "entities": intent_result.get("entities", {}), - "confidence": intent_result.get("confidence"), - "explanation": intent_result.get("explanation"), - } - ] - - primary_intent = intents[0] - return IntentDetectionResponse( - intent=primary_intent.get("intent", "general_query"), - entities=primary_intent.get("entities", {}), - confidence=primary_intent.get("confidence"), - explanation=primary_intent.get("explanation"), - # Store all intents for multi-intent processing - additional_data={"all_intents": intents}, - ) - except json.JSONDecodeError: - print(response) - # Fallback parsing if LLM doesn't return valid JSON - return self._fallback_intent_detection(request.user_message) - - except Exception as e: - self.error(f"Error detecting intent: {e}") - # Return fallback intent - return IntentDetectionResponse(intent="general_query", entities={}, explanation=f"Error in intent detection: {str(e)}") - - async def generate_followup_message(self, user_message: str, agent: Any, knowledge_topics: list[str]) -> str: - """Generate a contextually aware, empathetic follow-up message for the smart chat flow.""" - agent_name = getattr(agent, "name", None) or (agent.get("name") if isinstance(agent, dict) else None) or "your agent" - agent_config = getattr(agent, "config", None) or (agent.get("config") if isinstance(agent, dict) else None) or {} - domain = agent_config.get("domain", "") - recent_topics = knowledge_topics[-2:] if len(knowledge_topics) > 1 else knowledge_topics # Last 2 topics - - # Determine user's progress stage for empathetic response - progress_stage = "starting" if len(knowledge_topics) < 3 else "developing" if len(knowledge_topics) < 8 else "advanced" - - # Build contextual prompt with empathy - context_prompt = f""" -User just said: "{user_message}" -Agent name: {agent_name} -Agent domain: {domain or "not set yet"} -Recent topics added: {", ".join(recent_topics) if recent_topics else "none yet"} -Progress stage: {progress_stage} - -Generate a supportive follow-up message that: -1. Acknowledges what they just accomplished -2. Asks ONE helpful next step question (20-30 words) -3. Shows understanding of their agent-building journey -4. Relates to their specific domain/topics when possible - -Be encouraging and specific to their context. -""" - - llm_request = BaseRequest( - arguments={ - "messages": [ - { - "role": "system", - "content": "You are an encouraging agent-building coach. Acknowledge progress, then ask one specific, helpful question about their next step.", - }, - {"role": "user", "content": context_prompt}, - ], - "temperature": 0.5, - "max_tokens": 80, - } - ) - try: - response = await self.llm.query(llm_request) - content = response.content - if isinstance(content, str): - return content.strip() - elif isinstance(content, dict): - # Some LLMs return {"choices": [{"message": {"content": ...}}]} - try: - return content["choices"][0]["message"]["content"].strip() - except Exception: - return str(content) - else: - return str(content) - except Exception as e: - self.error(f"Error generating follow-up message: {e}") - # Return contextual fallback messages - if not knowledge_topics: - return f"Great start! What domain would you like {agent_name} to specialize in?" - elif len(knowledge_topics) < 3: - return f"Nice work building {agent_name}'s knowledge! What related topic should we add next?" - else: - return f"Your {domain or 'agent'} is looking good! What aspect would you like to deepen?" - - def _build_intent_detection_prompt( - self, user_message: str, chat_history: list[MessageData], domain_tree: DomainKnowledgeTree | None - ) -> str: - """Build the LLM prompt for intent detection.""" - # Convert domain tree to JSON for context - tree_json = "null" - if domain_tree: - try: - tree_json = yaml.safe_dump(domain_tree.model_dump(), sort_keys=False).replace("children: []", "") - except Exception: - tree_json = "null" - # Build chat history context - history_context = "" - if chat_history: - recent_messages = chat_history[-3:] # Only include recent context - history_context = "\n".join([f"{msg.role}: {msg.content}" for msg in recent_messages]) - prompt = f""" -You are an assistant in charge of managing an agentβs profile **and** its hierarchical domain-knowledge tree. - -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -TASK -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -1. **Intent Extraction** β Detect **every** intent in the userβs latest message. -2. **Entity & Instruction Extraction** β Pull any relevant entities (knowledge_path for tree navigation, name, domain, topics for agent specialties, tasks for agent responsibilities) and, for an `instruct` intent, capture the full instruction text. -3. **Path Construction** β For each new topic, return the **exact path** that already exists in - `tree_json`; append only the truly new node(s). - -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -AVAILABLE INTENTS -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -β’ `add_information` β user adds a new topic / knowledge area -β’ `remove_information` β user wants to remove/delete a topic from the knowledge tree -β’ `refresh_domain_knowledge` β user wants to rebuild / reorganize the tree -β’ `update_agent_properties` β user changes agent name, domain, topics, tasks -β’ `instruct` β user issues a command **about a specific topic's content** -β’ `general_query` β any other question or request - -A single message may contain multiple intents. - -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -INPUT VARIABLES -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -β’ `history_context` β recent chat (plain text) -β’ `tree_json` β **current** knowledge tree (YAML-like dict; see example) -β’ `user_message` β latest user utterance (plain text) - -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -RULES -ββββββββββββββββββββββββββββββββββββββββββββββββββββββββ -1. **Traverse the tree** - β’ Treat each `topic` in `tree_json` as one node. - β’ Find the deepest existing node(s) that match the userβs requested topic - (case-insensitive, ignore punctuation). - β’ Only create **new** node(s) for the missing remainder of the path. - β’ The returned `knowledge_path` list MUST start with `"root"` and follow the - *exact* topic names found in `tree_json`, preserving capitalization and spacing. - -2. **No duplicate branches** - β’ If the topic already exists anywhere in the tree, point to that exact path; - do **not** create a parallel branch. - β’ Search the entire tree structure (not just immediate children) for existing topics. - β’ Use case-insensitive matching to find existing topics. - -3. **Coupled updates** - β’ If the user wants the agent to *gain expertise* (topics or tasks) - **and** add that topic to knowledge, output **two** intents: - `update_agent_properties` **and** `add_information`. - β’ `instruct` is **never coupled** with any other intent. - -4. **`instruct` specifics** - β’ Choose the most relevant existing `knowledge_path`; create a new branch only if the subject is absent. - β’ Add an `"instruction_text"` field that contains the userβs command verbatim (trim greetings/pleasantries). - β’ Do **not** modify agent properties when handling `instruct`. - -5. **Entity heuristics** - β’ **Domain** β patterns like "be a[n]
- 
-
-Copyright Β© 2024 Aitomatic, Inc. Licensed under the MIT License.
-
-https://aitomatic.com
-
-
-
-Copyright Β© 2024 Aitomatic, Inc. Licensed under the MIT License.
-
-https://aitomatic.com
-
-
-
- {isChatPage && (
-
- )}
- {getPageTitle()}
-
- {isChatPage && agent_id && (
-
-
-
- )}
- {isChatPage && agent_id && isNaN(Number(agent_id)) && prebuiltAgent && (
-
-
-
- )}
-
-
- );
-};
-
-export default VersionNotification;
diff --git a/dana/contrib/ui/src/main.tsx b/dana/contrib/ui/src/main.tsx
deleted file mode 100644
index 16fd510da..000000000
--- a/dana/contrib/ui/src/main.tsx
+++ /dev/null
@@ -1,172 +0,0 @@
-import { StrictMode } from 'react';
-import { createRoot } from 'react-dom/client';
-import { BrowserRouter, Routes, Route, Navigate } from 'react-router-dom';
-import { Toaster } from 'sonner';
-import { Layout } from './components/layout';
-import AgentsPage from './pages/Agents';
-import AgentDetailPage from './pages/Agents/detail';
-import LibraryPage from './pages/Library';
-import DocumentationPage from './pages/Documentation';
-import SupportPage from './pages/Support';
-import StyleGuidePage from './pages/StyleGuide';
-import './index.css';
-import AgentChat from './pages/Agents/chat';
-import { analytics } from './lib/analytics';
-
-// Initialize Google Analytics
-analytics.initialize();
-
-// Initialize session tracking
-analytics.initializeSession();
-
-createRoot(document.getElementById('root')!).render(
-
-
-
-
-
-
- {/* Skeleton loader for agent detail */}
-
- );
- }
-
- if (error || (!isLoading && !selectedAgent)) {
- return (
-
-
-
-
-
-
-
-
-
-
-
-
- );
- }
-
- // --- Step 2: Training view ---
- return (
-
-
- Agent Not Found
-- {error || "The agent you're looking for doesn't exist or has been removed."} -
- -
- setShowComparison(false)}
- />
-
- {/* Cancel Confirmation Dialog */}
-
-
- );
-}
diff --git a/dana/contrib/ui/src/pages/Agents/index.tsx b/dana/contrib/ui/src/pages/Agents/index.tsx
deleted file mode 100644
index 2384c0b7f..000000000
--- a/dana/contrib/ui/src/pages/Agents/index.tsx
+++ /dev/null
@@ -1,918 +0,0 @@
-/* eslint-disable react-hooks/exhaustive-deps */
-/* eslint-disable @typescript-eslint/no-explicit-any */
-import { useEffect, useState } from 'react';
-import { useNavigate, useSearchParams } from 'react-router-dom';
-import { useAgentStore } from '@/stores/agent-store';
-import { apiService } from '@/lib/api';
-import { MyAgentTab } from './MyAgentTab';
-import { ExploreTab } from './ExploreTab';
-import { ImportAgentDialog } from '@/components/import-agent-dialog';
-import { NavArrowDown, Plus, Import, Search } from 'iconoir-react';
-import {
- Dialog,
- DialogContent,
- DialogFooter,
- DialogHeader,
- DialogTitle,
-} from '@/components/ui/dialog';
-import { Button } from '@/components/ui/button';
-import {
- type AgentSuggestion,
- type BuildAgentFromSuggestionRequest,
- type WorkflowInfo,
-} from '@/lib/api';
-// Removed React Flow imports - using simple HTML/CSS layout instead
-
-const DOMAINS = ['All domains', 'Finance', 'Semiconductor', 'Sales', 'Engineering', 'Research'];
-
-// Simple Workflow Step Box Component
-const WorkflowStepBox: React.FC<{
- step: string;
- stepNumber: number;
- isLast: boolean;
-}> = ({ step, isLast }) => {
- const formattedStep = step
- .replace(/_/g, ' ')
- .replace(/([A-Z])/g, ' $1')
- .trim();
-
- return (
-
-
-
-
-
-
- );
-};
-
-// Example Box Component
-const ExampleBox: React.FC<{
- example: string;
- stepNumber: number;
- isLast: boolean;
-}> = ({ example, isLast }) => {
- return (
-
-
- {!isLast && (
-
- {formattedStep}
-
-
-
-
- )}
-
-
- );
-};
-
-// Toggle Button Component
-const ToggleButton: React.FC<{
- onClick: () => void;
- isActive: boolean;
-}> = ({ onClick, isActive }) => {
- return (
-
- );
-};
-
-// Simple Workflow Chart Component
-const SimpleWorkflowChart: React.FC<{
- workflow: { name: string; steps: string[] };
- methods: string[];
- showExamples: boolean;
- setShowExamples: (show: boolean) => void;
- agentKey?: string;
-}> = ({ workflow, showExamples, agentKey }) => {
- if (!workflow.steps || workflow.steps.length === 0) {
- return (
-
-
-
- {!isLast && (
-
-
-
- )}
-
- No workflow steps defined
-
- );
- }
-
- // Agent-specific examples based on agent key
- const getExamplesForAgent = (agentKey?: string) => {
- if (!agentKey) {
- // Default Q&A Agent examples
- return [
- "What's the best time to visit Japan for cherry blossoms",
- "Agent refines query: 'cherry blossom season Japan travel dates optimal timing'",
- 'Search across uploaded travel documents using refined keywords',
- 'The best time to visit Japan for cherry blossoms is typically late March to early April...',
- 'Task completed successfully with comprehensive answer provided',
- ];
- }
-
- const agentType = agentKey.toLowerCase();
-
- if (agentType.includes('jordan') || agentType.includes('operational')) {
- return [
- "User:Turn my notes from today's meeting into a task list.",
- 'Agent compresses β input: meeting notes; output: structured task list.',
- 'Execute Steps: 1. Read meeting notes β 2. Identify action items β 3. Format into clear to-do list.',
- "Here's your to-do list: [example to-do list]",
- ];
- } else if (agentType.includes('nova') || agentType.includes('autonomous')) {
- return [
- 'User: Optimize warehouse inventory management system',
- "Agent identifies issues: 'Low stock alerts, inefficient reorder points, manual processes'",
- 'Analyze inventory data, implement automated reorder system, update warehouse layout',
- 'System optimized: 30% reduction in stockouts, 25% cost savings, automated alerts active',
- 'Task completed: Inventory management system fully optimized and operational',
- ];
- } else if (agentType.includes('dana')) {
- return [
- 'User: Is it better to pay off my mortgage early or invest in stocks?',
- "Agent refines query: 'mortgage prepayment vs stock investment financial comparison risk return'",
- 'Query Document: Agent reviews uploaded financial reports, mortgage agreements, and investment guides.',
- 'Query Knowledge: Cross-checks with general financial principles and market data.',
- 'Prepaying a mortgage guarantees savings equal to your interest rate, while stock investing offers...',
- 'Task completed.',
- ];
- } else {
- // Default Q&A Agent examples
- return [
- "User:What's the best time to visit Japan for cherry blossoms",
- "Agent refines query: 'cherry blossom season Japan travel dates optimal timing'",
- 'Search across uploaded travel documents using refined keywords',
- 'The best time to visit Japan for cherry blossoms is typically late March to early April...',
- 'Task completed successfully with comprehensive answer provided',
- ];
- }
- };
-
- const examples = getExamplesForAgent(agentKey);
-
- // Create complete workflow with User query and Complete task
- const completeWorkflowSteps = ['User query', ...workflow.steps, 'Complete task'];
-
- return (
-
-
- );
-};
-
-// Tab configuration with URL-friendly identifiers
-const TAB_CONFIG = {
- explore: 'Explore',
- my: 'My Agent',
-} as const;
-
-type TabId = keyof typeof TAB_CONFIG;
-
-export default function AgentsPage() {
- const navigate = useNavigate();
- const [searchParams, setSearchParams] = useSearchParams();
- const { agents, fetchAgents } = useAgentStore();
- const [myAgentSearch, setMyAgentSearch] = useState('');
- const [exploreSearch, setExploreSearch] = useState('');
- const [selectedDomain, setSelectedDomain] = useState('All domains');
- const [creating] = useState(false);
- const [headerCollapsed, setHeaderCollapsed] = useState(false);
- const [showCreateAgentPopup, setShowCreateAgentPopup] = useState(false);
- const [userInput, setUserInput] = useState('');
- const [suggestions, setSuggestions] = useState
- {/* Left Column: Workflow Steps */}
-
-
-
-
- {/* Right Column: Examples */}
-
- {completeWorkflowSteps.map((step, index) => (
-
-
-
-
- {completeWorkflowSteps.map((_step, index) => (
-
-
-
- ))}
-
- {/* Hero Section with Animated Background */}
-
- {/* Animated Background Elements - Only show when expanded */}
- {!headerCollapsed && (
- <>
-
- {/* Grid Pattern Overlay */}
-
-
- )}
-
- {/* Main Content with Smooth Transitions */}
-
-
- {/* Content Section */}
-
- {/* Main Title with Enhanced Typography */}
-
-
- {/* Enhanced Floating Elements - Only show when expanded */}
- {!headerCollapsed && <>}
-
- {/* Compact Header with Highlights - Show when collapsed */}
-
-
-
- {/* Enhanced Subtitle */}
- - Dana -
-- Agent Studio -
-- The complete platform for{' '} - building, training, and deploying{' '} - Dana Expert Agents -
- - {/* Feature Cards with Better Design */} -
- {/* Agent Maker - Available Now */}
-
-
- {/* Enhanced CTA Buttons */}
-
-
-
- {/* Experience Learner - Coming Soon */}
-
-
-
- Agent Maker
-- Create Dana Expert Agents with domain expertise and learning capabilities -
-
-
-
- {/* App Generators - Coming Soon */}
-
-
- Coming Soon
-
-
-
-
-
- Experience Learner
-- Dana Expert Agents that evolve and improve through continuous learning and feedback -
-
-
-
-
- Coming Soon
-
-
-
-
-
- App Generators
-- Deploy Dana Expert Agents to web, iOS, and Android with built-in app generation -
-
-
-
-
-
-
- {/* Main Title with Gradient Highlight */}
-
- - Dana Agent Studio -
- - {/* Subtle Description */} -- The complete platform for building, training, and deploying Dana Expert Agents -
-
- {/* Dana Agent Maker Feature */}
-
- {/* Search and Navigation */}
-
- agent.name.toLowerCase().includes(myAgentSearch.toLowerCase()) ||
- (agent.description || '').toLowerCase().includes(myAgentSearch.toLowerCase()),
- )}
- navigate={navigate}
- handleCreateAgent={handleCreateAgent}
- creating={creating}
- onSwitchToPretrained={() => setActiveTab('explore')}
- />
- )}
- {activeTab === 'Explore' && (
-
-
- {/* Create Agent Popup */}
-
-
-
- {/* Enhanced Tabs */}
-
-
-
-
-
-
-
-
-
-
-
-
-
- {/* Tab Content */}
- {activeTab === 'My Agent' && (
-
-
- )}
-