⚙️ Configuration

co_ai uses Hydra for flexible and hierarchical configuration. Each agent, tool, and subsystem loads settings from a structured YAML config. This allows you to dynamically switch models, pipelines, memory backends, and more.

---

🗂️ Configuration Structure

The root config lives in config/config.yaml and includes nested paths for:

defaults:
  - db: default
  - logging: default
  - report: default
  - agents:
      - generation
      - ranking
      - review
      - reflection
      - evolution
      - proximity
      - meta_review
````

---

## 🧠 Memory Configuration (`db.yaml`)

Example for a PostgreSQL database with `pgvector`:

```yaml
host: localhost
port: 5432
database: co_ai
user: postgres
password: yourpassword

# Optional store overrides
extra_stores: []

---

📜 Logging (logging.yaml)

logger:
  log_path: logs/
  log_file: ""

• log_path: Base directory for logs
• log_file: Specific log file; if empty, will auto-generate using timestamp and run ID

---

📊 Reporting (report.yaml)

generate_report: true
format: yaml

---

🧪 Agent Configuration

Each agent has its own file under config/agents/. For example:

✨ Generation Agent

name: generation
prompt_type: file
prompt_file: generation_goal_aligned.txt
top_k: 5

🧠 Reflection Agent

name: reflection
prompt_type: file
prompt_file: reflect_consistency.txt
preferences:
  - goal_consistency
  - factual
  - reliable_source
  - simplicity

---

🧬 Prompt Tuning Agent

name: prompt_tuning
model: qwen2.5
num_prompts: 10

---

📁 Prompt Directory Override

You can override the prompt directory from the command line or within the context:

PROMPT_DIR: "prompts"

---

🧪 Example Run with Overrides

You can override any value from the CLI:

python co_ai/main.py goal="Can AI assist scientific discovery?" logging.logger.log_path=run_logs

Or define a full run config:

goal: "Evaluate LLMs for literature review"
run_id: "review_run_01"

---

📘 Tips

• Every config is passed to its agent as a flat dict
• Use .get() for optional config keys
• Log config values using OmegaConf.to_yaml(cfg)

---

🔁 Runtime Mutation

Agents can mutate config or context, but best practice is to treat config as static and context as dynamic.

---