Skip to content

Cortex

Cortex research subnet: Bounty and agentic Proof on Bittensor.

The Python implementation runs the gateway and both challenges on a master, verifies sealed rewards in independent validators, and isolates research work in Firecracker guests. Algorithm 2 assigns up to 30% of emission to Bounty and 70% to Proof; activating it requires a new owner-signed trust root. Proof uses Cortex's own recursive language-model engine with persistent memory, context compaction and bounded tool execution.

Cortex is experimental research software. Offline tests exercise submission, scoring, sealing and validator dispatch through fake external boundaries. A live model smoke is distinct from live KVM execution or confirmed on-chain payment.

Current launch mode

The initial production mode enables Bounty against the configured CortexLM/backend public feed and leaves Proof execution unwired. The signed trust root still contains bounty = 2000 and proof = 8000: Proof emits ChallengeInternal absences and its share burns to UID 0. Never renormalize Bounty to 100%. Production pairing, report intake and adjudication stay in CortexLM/backend; Cortex reads its immutable public scoring snapshots. Algorithm 2 pays one point per valid report, proportionally across authors; ten valid reports across expected participants unlock the full Bounty share. The unsigned 30/70 template changes nothing until the trust-root migration is completed on the gateway and validators.

Installation

Linux, Python 3.12 or 3.13, uv, and libsodium 1.0.18 or newer are required. The VM host additionally requires working /dev/kvm, Firecracker and jailer.

sudo apt-get install libsodium23
uv sync --locked --extra chain --group build
uv run cortex --help

Usage

uv run cortex master --help
uv run cortex validator --help
uv run cortex vm-host --help
uv run cortex miner --help
uv run cortex topic-create --help

Configure the master using private credential files and signed trust roots. An operator gives Proof a research objective; its topic agent prepares an experiment, measures a baseline and proposes rules and miner documentation. A topic opens only after the control plane verifies the host's execution receipts and signs the resulting document. No research task catalog is built in.

Development

uv run ruff format --check src tests scripts
uv run ruff check src tests scripts
uv run mypy
uv run pytest -m 'not live'
uv run python scripts/check_deploy.py --check-examples
uv build --no-build-isolation

Tests cover signatures and Rust wire vectors, replay protection, feed outages, artifact validation, topic setup, rejected submissions, VM lifecycle failures, RLM recursion/compaction, reward allocation and sealed-weight submission. CI runs offline and never rents a GPU or boots Firecracker.

Contributing

Keep research content out of source control. Add tests for observable behavior at service boundaries, and update the matching miner documentation when an API changes. Follow the agent contract and the PR review requirements.

License

Apache-2.0.

About

Autonomous research network on Bittensor subnet 100, turning reproducible AI research into shared knowledge.

Resources

Code of conduct

Contributing

Security policy

Stars

159 stars

Watchers

1 watching

Forks

Releases

Packages

Used by

Contributors

Languages