Metadata-Version: 2.4
Name: emocore
Version: 0.5.2
Summary: A runtime governance layer that enforces hard behavioral bounds in autonomous agents.
Author: Sarthaksahu777
License: MIT
Project-URL: Homepage, https://github.com/Sarthaksahu777/Emocore
Project-URL: Repository, https://github.com/Sarthaksahu777/Emocore
Project-URL: Documentation, https://github.com/Sarthaksahu777/Emocore#readme
Keywords: agents,runtime,governance,control,AI,LLM,safety
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3 :: Only
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy
Provides-Extra: dev
Requires-Dist: pytest>=7.0; extra == "dev"
Requires-Dist: pytest-cov; extra == "dev"
Dynamic: license-file

# EmoCore

![PyPI version](https://img.shields.io/pypi/v/emocore)
![License](https://img.shields.io/github/license/Sarthaksahu777/Emocore)

EmoCore is a runtime governor that forces autonomous systems to halt when they stop making progress.

Modern agents can reason, plan, and retry — but they do not know when to stop.
EmoCore adds a deterministic execution-time control layer that guarantees finite-time halting under sustained non-progress, exhaustion, or unsafe conditions.

This is not a policy.
This is not alignment.
This is a hard execution boundary.

---

## Why EmoCore Exists

Autonomous systems fail structurally, not cognitively.

In production, engineers observe:
- Agents retrying the same action forever
- Multi-agent systems deadlocking but continuing to run
- Urgency causing reckless escalation instead of stopping
- External timeouts killing processes without understanding internal state
- Silent degradation (“zombie agents”) that keep consuming tokens, time, or compute

Policy-level intelligence cannot enforce bounded execution.

EmoCore exists to solve one problem only:
> Guarantee that an autonomous system halts in finite time when progress stalls.

---

## What EmoCore Does (At Runtime)

EmoCore sits between your agent loop and execution.

It does not choose actions.
It does not inspect reasoning.
It decides whether the system is still allowed to act.

Conceptually:
```text
┌────────────────────────┐
│     Environment        │
│  (reward, urgency…)    │
└───────────┬────────────┘
            ↓
┌────────────────────────┐
│   Agent / LLM /        │
│   Planner / Controller │
│   (chooses actions)    │
└───────────┬────────────┘
            ↓
┌────────────────────────┐
│      EmoCore           │
│  Runtime Governor      │
│  • Tracks stress       │
│  • Enforces budgets    │
│  • Decides HALT / GO   │
└───────────┬────────────┘
            ↓
┌────────────────────────┐
│   Action Execution     │
│   (tools, APIs, motors)│
└────────────────────────┘
```

Execution is no longer “run unless killed”. Execution becomes permissioned.

---

## Observable Behavior (Not Philosophy)

EmoCore provides mechanical guarantees you can see happen:

1. **Deterministic halting** under zero or negative progress
2. **Explicit failure modes**:
   - `EXHAUSTION`: Budget depleted due to sustained effort without recovery.
   - `STAGNATION`: Progress stalled for longer than `stagnation_window`.
   - `OVERRISK`: Risk budget exceeded safety threshold.
   - `SAFETY`: Exploration budget exceeded limits.
   - `EXTERNAL`: Hard step limit reached (`max_steps`).

3. **No learning, no randomness, no drift**:
   - Identical governance behavior across agents and models.
   - Irreversible halting (no silent recovery).
   - If the same signals are fed in, the same halt always occurs.

---

## What EmoCore Is NOT

This is critical. EmoCore is not:

❌ A policy or planner
❌ Reinforcement learning
❌ Alignment or preference modeling
❌ Emotion simulation
❌ A model wrapper or prompt technique
❌ A replacement for agent logic

> EmoCore does not influence what the system thinks — it only enforces whether it may continue acting.

---

## How EmoCore Works (Mental Model)

EmoCore maintains two internal layers:

### 1. Pressure (Unbounded)
Pressure represents accumulated execution stress:
- Repeated failure
- Lack of progress
- Urgency
- Difficulty

Pressure accumulates without bound via the `integrate()` function.

### 2. Budgets (Bounded)
Budgets represent permission to act:
- **Effort**: Energy available for action.
- **Persistence**: Wille to continue despite failure.
- **Risk**: Tolerance for unsafe actions.
- **Exploration**: Allowance for novelty seeking.

Budgets are strictly bounded in `[0, 1]` via clamping.

```text
Pressure (Unbounded)              Budgets (Bounded)
──────────────────────            ───────────────────
Frustration   ↑↑↑↑↑               Effort        ┌───┐
Urgency       ↑↑↑                 Persistence   │   │
Difficulty    ↑↑                  Risk          │   │
                                  Exploration   └───┘
     (accumulates forever)        (clipped to [0,1])

          ↑                               ↓
   Stress increases            Permission collapses
                               ↓
                           TERMINAL HALT
```

> **Key asymmetry**: Pressure can grow forever. Permission cannot.

Under sustained stress, permission always collapses.

### 3. Dynamic Recovery
Before halting, EmoCore attempts to recover.
If pressure lifts, the system enters `RECOVERING` mode:
- Effort and persistence regenerate.
- Risk is frozen (safety interlock).
- If stress returns before full recovery, exhaustion accelerates.

This mimics biological "second wind" mechanics — but deterministically.

---

## Failure Semantics (What Happens When Things Go Wrong)

EmoCore checks failure conditions every step, in priority order (defined in `engine.py`):

| Priority | Condition | Failure Type | Result |
|:---|:---|:---|:---|
| 1 | Exploration > max_exploration | `SAFETY` | Halts |
| 2 | Risk > max_risk | `OVERRISK` | Halts |
| 3 | Effort ≤ exhaustion_threshold | `EXHAUSTION` | Halts |
| 4 | Stagnating & Low Effort | `STAGNATION` | Halts |
| 5 | Steps ≥ max_steps | `EXTERNAL` | Halts |

> **When EmoCore halts, the system is done unless externally reset.**

This is fail-closed by design.

---

## Installation

### Option 1: Local Development Install

```bash
git clone https://github.com/Sarthaksahu777/Emocore
cd Emocore
pip install -e .
```

> [!IMPORTANT]
> **Implementation Note**: The PyPI distribution name is `emocore`, but the implementation package is named `core`.
> - **Install**: `pip install emocore` (or `pip install -e .` locally)
> - **Import**: `from core import EmoCoreAgent`, `from core import step`, etc.

### Option 2: Configuration
You can select different governance profiles at initialization:

```python
from core import EmoCoreAgent, PROFILES, ProfileType

# 1. Balanced (Default): Good for general tasks
agent = EmoCoreAgent()

# 2. Conservative: Halts early on risk/uncertainty
agent = EmoCoreAgent(PROFILES[ProfileType.CONSERVATIVE])

# 3. Aggressive: Tolerates higher risk/exhaustion
agent = EmoCoreAgent(PROFILES[ProfileType.AGGRESSIVE])
```

---

## API Reference: Inputs

The `step()` function accepts three normalized signals `[0.0, 1.0]`:

| Signal | Meaning | Effect on Governance |
|:---|:---|:---|
| `reward` | Progress towards goal | High reward maintains **Effort/Persistence**. Zero reward generally triggers **Stagnation**. |
| `novelty` | New/unexpected info | High novelty boosts **Curiosity/Exploration**. |
| `urgency` | Time/resource pressure | High urgency burns **Effort** faster but boosts short-term **Risk Tolerance**. |

---

## Minimal Example (60-Second Test)

```python
from core import EmoCoreAgent, step, Signals

# Initialize with default "BALANCED" profile
agent = EmoCoreAgent()

print("Agent started.")

while True:
    # 1. Feed signals from the environment
    # reward=0.0 simulates no progress
    result = step(agent, Signals(reward=0.0, novelty=0.0, urgency=0.1))
    
    # 2. Check strict halting condition
    if result.halted:
        print(f"Agent halted! Reason: {result.reason} ({result.failure})")
        break
        
    print(f"Step {agent.engine.step_count}: Effort={result.budget.effort:.2f}")

# Result: The system halts deterministically in finite steps.
```

**Result:**
The system halts deterministically.
- No timeouts.
- No heuristics.
- No model cooperation required.

---

## Guarantees (What You Can Rely On)

EmoCore guarantees:
- **Finite-time halting** under sustained non-progress.
- **Deterministic behavior**: Same history = same outcome (using fixed matrices).
- **Irreversible halting**: Once halted, budget remains zeroed (`guarantees.py`).
- **Model-agnostic control**: Works with any agent architecture.
- **No silent degradation**: Failures are explicit and typed.

These are execution guarantees, not performance claims.

---

## Who Should Use EmoCore

**Use EmoCore if:**
- Your system runs for many steps.
- Retries automatically.
- Uses tools or APIs.
- Consumes tokens, time, or compute.
- Must not run forever.

**Do not use EmoCore if:**
- You are doing single-shot prompting.
- You want learning or optimization.
- You expect graceful degradation instead of hard stops.

---

## Roadmap

- **v1.0 (Planned)**: Stable install + signal extraction definitions.
- **v1.1**: Adapters for agent frameworks.
- **v2.x**: Adaptive profiles (optional, non-learning).

---

## Documentation Suite

For deep details, see the `docs/` directory:

- [**Technical Definition**](docs/DEFINITION.md) — Core properties and fundamental logic.
- [**Architecture & Lifecycle**](docs/ARCHITECTURE.md) — Component responsibilities and terminality invariants.
- [**Failure Modes**](docs/FAILURE_MODES.md) — Specification of `EXHAUSTION`, `OVERRISK`, and `STAGNATION`.
- [**Terminology**](docs/TERMINOLOGY.md) — Definitive meanings for data primitives and system states.

---

## One-Line Summary

> EmoCore enforces bounded agency by guaranteeing finite-time halting for autonomous systems — regardless of how the model behaves.
