# MONIKA

MONIKA is a local MCP-serving runtime with a Proto language model, a baseline
training entry point, and a small set of maintenance utilities.

## Active Paths

- `salience_os_seed/runtime/` - Core runtime, MCP bridge, orchestrator, and driver.
- `salience_os_seed/proto_lm/` - ProtoLM implementations and checkpoint management.
- `start_mcp_server.py` - Root CLI entry point for running MONIKA as an MCP server.
- `start.standard.py` - Synthetic baseline ProtoLM training harness.
- `attic/` - Archived legacy scripts, manifests, and older checkpoints.
- `attic_sweep.tmp.py` - Cleanup script for archiving temporary helpers.

## Launch the MCP Server

Run from the repository root:

```bash
python start_mcp_server.py
```

Optional arguments let you load saved state or a specific checkpoint:

```bash
python start_mcp_server.py --state storage/conversation_state.json --checkpoint storage/proto_lm/checkpoint.pt
```

## Train the ProtoLM Baseline

Run the baseline training harness from the repository root:

```bash
python start.standard.py
```

This uses the synthetic baseline corpus at
`data/local_benchmarks/synthetic_baseline_corpus.txt` and writes the baseline
checkpoint to `storage/proto_lm/synthetic_baseline.pt`.

## Run Tests

Pytest coverage lives under `tests/`.

```bash
pytest
```

## Use `attic_sweep.tmp.py`

Archive temporary helpers and extra maintenance artifacts with:

```bash
python attic_sweep.tmp.py
```

The sweep preserves files under `attic/` while moving:

- `*.tmp.py` files, except `attic_sweep.tmp.py`
- Legacy maintenance scripts listed by the sweeper
- Extra non-canonical checkpoints listed by the sweeper