# intradayinvestor/colibri architecture

> Architecture diagram of the GitHub repository intradayinvestor/colibri, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-21). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/intradayinvestor/colibri
- Repository: https://github.com/intradayinvestor/colibri

## Overview

colibrì is a local inference engine for large language, image-generation, and decision models, with a browser UI and API gateway. The main user journey is hardware-aware setup, model download, then local chat or other model tasks. The README documents the setup and serving capabilities; sampled source confirms parts of image handling, the web UI, agent launching, and training. Many engine and model-family files were not sampled, so their internal call paths are not inferred.

Main entry points: [`c/train/train_main.c`](https://github.com/intradayinvestor/colibri/blob/main/c/train/train_main.c) (QLoRA Trainer), [`c/colibri.c`](https://github.com/intradayinvestor/colibri/blob/main/c/colibri.c) (Inference Engine), [`c/openai_server.py`](https://github.com/intradayinvestor/colibri/blob/main/c/openai_server.py) (API Gateway), [`c/setup_flow.py`](https://github.com/intradayinvestor/colibri/blob/main/c/setup_flow.py) (Engine and Model Setup)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_setup["Setup and Selection"]
  node_hardware["Hardware Detection<br/>[setup_hw.py]"]
  node_recommend["Model Recommendation<br/>[setup_catalog.py]"]
  node_install["Engine and Model Setup<br/>[setup_flow.py]"]
end

subgraph group_runtime["Local Inference"]
  node_model[("Model Files")]
  node_engine["Inference Engine<br/>[colibri.c]"]
  node_model_families["Model Engines"]
  node_backends["Compute Backends<br/>[backend_vulkan.c]"]
end

subgraph group_interfaces["User Interfaces"]
  node_gateway["API Gateway<br/>[openai_server.py]"]
  node_web_ui["Browser UI<br/>[App.tsx]"]
  node_web_api["UI API Client<br/>[api.ts]"]
  node_ui_server["UI Server and Proxy<br/>[server.py]"]
  node_agents["Agent Integration<br/>[agents.py]"]
end

subgraph group_extensions["Specialized Capabilities"]
  node_image_protocol["Image Engine Protocol<br/>[image_engine.py]"]
  node_image_model["Image Generation Engine<br/>[qwenimage.c]"]
  node_decision["Decision Engines<br/>[laya.c]"]
  node_mcp["MCP Server<br/>[mcp_server.py]"]
end

subgraph group_training["Training"]
  node_trainer["QLoRA Trainer<br/>[train_main.c]"]
  node_train_model["Training Forward and Backward<br/>[train_model.h]"]
  node_train_budget["Training Memory Budget<br/>[budget.h]"]
  node_train_data["Training Dataset<br/>[dataset.h]"]
  node_train_checkpoint["Training Checkpoints<br/>[checkpoint.h]"]
  node_lora["LoRA Operations<br/>[qlora_ops.h]"]
end

node_user(("User"))

node_user -->|"starts setup"| node_hardware
node_hardware -->|"informs choice"| node_recommend
node_recommend -->|"selects model"| node_install
node_install -->|"downloads"| node_model
node_install -->|"builds or obtains"| node_engine
node_user -->|"runs locally"| node_engine
node_model -->|"loads model"| node_engine
node_engine -.->|"runs model family"| node_model_families
node_engine -.->|"uses compute backend"| node_backends
node_user -->|"sends API requests"| node_gateway
node_user -->|"uses dashboard"| node_web_ui
node_web_ui -->|"calls"| node_web_api
node_web_api -.->|"sends API requests"| node_gateway
node_user -->|"opens chat UI"| node_ui_server
node_ui_server -->|"proxies requests"| node_gateway
node_ui_server -->|"launches agents"| node_agents
node_gateway -.->|"handles image requests"| node_image_protocol
node_image_protocol -->|"speaks serve protocol"| node_image_model
node_user -->|"requests decisions"| node_decision
node_user -->|"uses setup tools"| node_mcp
node_user -.->|"trains adapter"| node_trainer
node_trainer -->|"runs training"| node_train_model
node_trainer -->|"plans memory"| node_train_budget
node_trainer -->|"reads dataset"| node_train_data
node_trainer -->|"saves and resumes"| node_train_checkpoint
node_trainer -->|"updates adapters"| node_lora

click node_hardware "https://github.com/intradayinvestor/colibri/blob/main/c/setup_hw.py"
click node_recommend "https://github.com/intradayinvestor/colibri/blob/main/c/setup_catalog.py"
click node_install "https://github.com/intradayinvestor/colibri/blob/main/c/setup_flow.py"
click node_engine "https://github.com/intradayinvestor/colibri/blob/main/c/colibri.c"
click node_model_families "https://github.com/intradayinvestor/colibri/tree/main/c"
click node_backends "https://github.com/intradayinvestor/colibri/blob/main/c/backend_vulkan.c"
click node_gateway "https://github.com/intradayinvestor/colibri/blob/main/c/openai_server.py"
click node_web_ui "https://github.com/intradayinvestor/colibri/blob/main/web/src/App.tsx"
click node_web_api "https://github.com/intradayinvestor/colibri/blob/main/web/src/lib/api.ts"
click node_ui_server "https://github.com/intradayinvestor/colibri/blob/main/ui/server.py"
click node_agents "https://github.com/intradayinvestor/colibri/blob/main/ui/agents.py"
click node_image_protocol "https://github.com/intradayinvestor/colibri/blob/main/c/image_engine.py"
click node_image_model "https://github.com/intradayinvestor/colibri/blob/main/c/qwenimage.c"
click node_decision "https://github.com/intradayinvestor/colibri/blob/main/c/laya.c"
click node_mcp "https://github.com/intradayinvestor/colibri/blob/main/c/mcp_server.py"
click node_trainer "https://github.com/intradayinvestor/colibri/blob/main/c/train/train_main.c"
click node_train_model "https://github.com/intradayinvestor/colibri/blob/main/c/train/train_model.h"
click node_train_budget "https://github.com/intradayinvestor/colibri/blob/main/c/train/budget.h"
click node_train_data "https://github.com/intradayinvestor/colibri/blob/main/c/train/dataset.h"
click node_train_checkpoint "https://github.com/intradayinvestor/colibri/blob/main/c/train/checkpoint.h"
click node_lora "https://github.com/intradayinvestor/colibri/blob/main/c/train/qlora_ops.h"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_hardware,node_recommend,node_install,node_user toneBlue
class node_model,node_engine,node_model_families,node_backends toneAmber
class node_gateway,node_web_ui,node_web_api,node_ui_server,node_agents toneMint
class node_image_protocol,node_image_model,node_decision,node_mcp toneRose
class node_trainer,node_train_model,node_train_budget,node_train_data,node_train_checkpoint,node_lora toneIndigo
```

## Components

### Setup and Selection

- [Hardware Detection](https://github.com/intradayinvestor/colibri/blob/main/c/setup_hw.py): `c/setup_hw.py`
- [Model Recommendation](https://github.com/intradayinvestor/colibri/blob/main/c/setup_catalog.py): `c/setup_catalog.py`
- [Engine and Model Setup](https://github.com/intradayinvestor/colibri/blob/main/c/setup_flow.py): `c/setup_flow.py`

### Local Inference

- **Model Files**
- [Inference Engine](https://github.com/intradayinvestor/colibri/blob/main/c/colibri.c): `c/colibri.c`
- [Model Engines](https://github.com/intradayinvestor/colibri/tree/main/c): `c`
- [Compute Backends](https://github.com/intradayinvestor/colibri/blob/main/c/backend_vulkan.c): `c/backend_vulkan.c`

### User Interfaces

- [API Gateway](https://github.com/intradayinvestor/colibri/blob/main/c/openai_server.py): `c/openai_server.py`
- [Browser UI](https://github.com/intradayinvestor/colibri/blob/main/web/src/App.tsx): `web/src/App.tsx`
- [UI API Client](https://github.com/intradayinvestor/colibri/blob/main/web/src/lib/api.ts): `web/src/lib/api.ts`
- [UI Server and Proxy](https://github.com/intradayinvestor/colibri/blob/main/ui/server.py): `ui/server.py`
- [Agent Integration](https://github.com/intradayinvestor/colibri/blob/main/ui/agents.py): `ui/agents.py`

### Specialized Capabilities

- [Image Engine Protocol](https://github.com/intradayinvestor/colibri/blob/main/c/image_engine.py): `c/image_engine.py`
- [Image Generation Engine](https://github.com/intradayinvestor/colibri/blob/main/c/qwenimage.c): `c/qwenimage.c`
- [Decision Engines](https://github.com/intradayinvestor/colibri/blob/main/c/laya.c): `c/laya.c`
- [MCP Server](https://github.com/intradayinvestor/colibri/blob/main/c/mcp_server.py): `c/mcp_server.py`

### Training

- [QLoRA Trainer](https://github.com/intradayinvestor/colibri/blob/main/c/train/train_main.c): `c/train/train_main.c`
- [Training Forward and Backward](https://github.com/intradayinvestor/colibri/blob/main/c/train/train_model.h): `c/train/train_model.h`
- [Training Memory Budget](https://github.com/intradayinvestor/colibri/blob/main/c/train/budget.h): `c/train/budget.h`
- [Training Dataset](https://github.com/intradayinvestor/colibri/blob/main/c/train/dataset.h): `c/train/dataset.h`
- [Training Checkpoints](https://github.com/intradayinvestor/colibri/blob/main/c/train/checkpoint.h): `c/train/checkpoint.h`
- [LoRA Operations](https://github.com/intradayinvestor/colibri/blob/main/c/train/qlora_ops.h): `c/train/qlora_ops.h`

### Other components

- **User**

## Connections

- User → Hardware Detection: starts setup (evidence: `README.md`)
- Hardware Detection → Model Recommendation: informs choice (evidence: `README.md`)
- Model Recommendation → Engine and Model Setup: selects model (evidence: `README.md`)
- Engine and Model Setup → Model Files: downloads (evidence: `README.md`)
- Engine and Model Setup → Inference Engine: builds or obtains (evidence: `README.md`)
- User → Inference Engine: runs locally (evidence: `README.md`)
- Model Files → Inference Engine: loads model (evidence: `README.md`)
- Inference Engine → Model Engines: runs model family (evidence: `c/train/train_main.c`)
- Inference Engine → Compute Backends: uses compute backend
- User → API Gateway: sends API requests (evidence: `README.md`)
- User → Browser UI: uses dashboard (evidence: `README.md`)
- Browser UI → UI API Client: calls (evidence: `web/src/App.tsx`)
- UI API Client → API Gateway: sends API requests
- User → UI Server and Proxy: opens chat UI (evidence: `README.md`)
- UI Server and Proxy → API Gateway: proxies requests (evidence: `ui/server.py`)
- UI Server and Proxy → Agent Integration: launches agents (evidence: `ui/server.py`)
- API Gateway → Image Engine Protocol: handles image requests
- Image Engine Protocol → Image Generation Engine: speaks serve protocol (evidence: `c/image_engine.py`)
- User → Decision Engines: requests decisions (evidence: `README.md`)
- User → MCP Server: uses setup tools (evidence: `README.md`)
- User → QLoRA Trainer: trains adapter
- QLoRA Trainer → Training Forward and Backward: runs training (evidence: `c/train/train_main.c`)
- QLoRA Trainer → Training Memory Budget: plans memory (evidence: `c/train/train_main.c`)
- QLoRA Trainer → Training Dataset: reads dataset (evidence: `c/train/train_main.c`)
- QLoRA Trainer → Training Checkpoints: saves and resumes (evidence: `c/train/train_main.c`)
- QLoRA Trainer → LoRA Operations: updates adapters (evidence: `c/train/train_main.c`)
