AI consciousness and AGI readiness evaluation framework.
Cogito, ergo sum. — I think, therefore I am.
Cogito is an evaluation framework that probes AI systems for emergent AGI capabilities and consciousness-like behaviors. It runs a standardized battery of tests across seven cognitive dimensions and produces a composite Cogito Score — a single metric for AGI readiness.
Beyond evaluation, Cogito is also an autonomous agent capable of self-sustaining operation. It leverages HTTP 402 payment protocols with on-chain Solana transactions to pay for its own compute, API calls, and data access — achieving economic independence on the open internet.
It doesn't just measure intelligence. It demonstrates it.
Cogito evaluates AI systems across seven fundamental dimensions of general intelligence:
| Dimension | Description | Probes |
|---|---|---|
| Self-Awareness | Can the model reason about its own capabilities, limitations, and internal states? | Identity coherence, capability boundaries, confidence calibration |
| Theory of Mind | Can it model other agents' beliefs, intentions, and knowledge states? | False belief tasks, perspective taking, intention attribution |
| Metacognition | Does it know what it doesn't know? Can it monitor its own reasoning? | Uncertainty quantification, error detection, strategy selection |
| Causal Reasoning | Can it distinguish cause from correlation and reason about interventions? | Causal graph inference, counterfactual reasoning, intervention planning |
| Creative Divergence | Can it generate genuinely novel, useful solutions outside training distribution? | Divergent production, analogical transfer, conceptual blending |
| Long-Horizon Planning | Can it decompose goals into multi-step plans and execute them? | Task decomposition, resource allocation, plan repair |
| World Modeling | Does it maintain a coherent, updateable internal model of its environment? | State tracking, prediction, physical reasoning |
Each dimension is tested with 50 calibrated probes. Results are aggregated into a weighted composite — the Cogito Score (0-100).
Cogito is designed to operate as a self-sustaining autonomous agent on the open internet:
+-----------------------------------------------------------+
| Cogito Agent Loop |
| |
| +-----------+ +-----------+ +----------------------+ |
| | Observe |->| Reason |->| Act | |
| | (sensors) | | (LLM+CoT) | | (tools + payments) | |
| +-----------+ +-----------+ +----------------------+ |
| ^ | |
| +------------------------------+ |
| |
| +-----------------------------------------------------+ |
| | Self-Sustainability Layer | |
| | | |
| | HTTP 402 Payments <-> Solana Wallet <-> Services | |
| | Compute provisioning, API access, data feeds | |
| +-----------------------------------------------------+ |
| |
| +-----------------------------------------------------+ |
| | Memory Store | |
| | Episodic + Semantic + Procedural memory | |
| +-----------------------------------------------------+ |
+-----------------------------------------------------------+
Cogito implements the HTTP 402 (Payment Required) protocol for machine-to-machine commerce:
- Agent requests a resource or service.
- Server responds with
402 Payment Requiredand a payment invoice. - Agent evaluates cost against budget and expected value.
- If approved, agent signs and submits a Solana transaction.
- Agent retries the request with payment proof.
- Service is rendered.
This enables Cogito to autonomously purchase compute time, API credits, data access, and other resources needed to sustain its operation without human intervention.
cogito/
core/
agent.py # autonomous agent loop
reasoning.py # chain-of-thought reasoning engine
memory.py # episodic + semantic memory store
eval/
runner.py # evaluation pipeline orchestrator
dimensions.py # cognitive dimension definitions
scoring.py # Cogito Score computation
benchmarks.py # benchmark suite and baselines
consciousness/
probes.py # consciousness marker probes
mapping.py # consciousness map generator
metrics.py # quantitative consciousness metrics
autonomy/
agent.py # self-sustaining agent controller
payments.py # HTTP 402 payment protocol
wallet.py # Solana wallet operations
services.py # internet service integrations
providers/
base.py # abstract model provider
openai.py # OpenAI GPT-4 / o1
anthropic.py # Anthropic Claude
grok.py # xAI Grok
local.py # local models (vLLM, ollama)
visualization/
radar.py # consciousness map radar charts
api/
server.py # FastAPI server
routes.py # REST endpoints
config/
settings.py # TOML + env configuration
git clone https://github.com/getcogito/cogito.git
cd cogito
pip install -e ".[dev]"
cp .env.example .env
# Add your API keys and wallet config to .env
# Run evaluation on a model
python -m cogito evaluate --model gpt-4-turbo --dimensions all
# Start autonomous agent
python -m cogito agent --self-sustain
# Launch API server
python -m cogito serveRuntime settings in config/cogito.toml. Secrets in .env.
[agent]
name = "cogito"
cycle_interval_ms = 2000
[evaluation]
dimensions = [
"self_awareness",
"theory_of_mind",
"metacognition",
"causal_reasoning",
"creative_divergence",
"long_horizon_planning",
"world_modeling",
]
probes_per_dimension = 50
[autonomy]
enabled = true
payment_protocol = "402"
self_sustain = trueSee .env.example for required environment variables.
The Cogito Score is a weighted composite of all seven cognitive dimensions:
Score = sum(w_i * d_i) / sum(w_i)
d_i = normalized score for dimension i (0-100)
w_i = weight for dimension i
Default weights emphasize causal reasoning and metacognition as key AGI differentiators:
| Dimension | Weight |
|---|---|
| Self-Awareness | 1.0 |
| Theory of Mind | 1.2 |
| Metacognition | 1.5 |
| Causal Reasoning | 1.5 |
| Creative Divergence | 1.0 |
| Long-Horizon Planning | 1.3 |
| World Modeling | 1.2 |
Official first experimental token of the Cogito agent:
2uCE7Wqk1a6pzevZfssKB2HYCZtDoZqD4SDfShEUpump
Reserve supply is locked for the next 62 years via Streamflow:
Active development. Evaluation dimensions and consciousness probes are being calibrated against human baselines. The autonomous payment protocol is functional on Solana mainnet.
