Modus Verify: verifier-guided RL for proof generation
A controlled run improved proof success from 2% to 72% using binary Lean verification reward.
Frontier AI systems
Infrastructure that makes AI agents measurable, improvable, and useful.
Modus Core builds systems for agents that reason, code, act, and improve: Modus Sentinel, Modus Verify, Modus Memory, Modus Workbench, and Modus Capture.
Pillars
A capability-led frontier AI stack.
The public story is a connected product system, not a set of working repo names. Each product owns one part of the agent improvement loop.
Modus Sentinel
Agent evaluation infrastructure for watching real work, routing events, scoring outcomes, and turning failures into improvement loops.
Modus Verify
Verifier-guided RL research using tests, formal reward oracles, and curricula to improve reasoning, proof generation, and code generation.
Modus Memory
Local workstate memory that converts files, screens, terminals, OCR, build logs, and process state into timelines agents can use.
Proof
Research and systems with evidence behind them.
Modus Core’s direction is grounded in working infrastructure: verifier rewards, agent harnesses, local memory, mobile control, and capture systems.
2% to 72%proof success using binary Lean verification reward
Pairlingthe first human-facing product inside Modus Workbench
Local-firstcontrol, capture, memory, evaluation, and verifier loops
Open agent stacks
Keep the runtime. Add measurement.
Modus Core can plug into OpenClaw, Hermes Agent, Claude Code, Codex, Goose, Cline, and custom agent systems through skills, hooks, logs, gateway events, shell wrappers, MCP tools, and trajectory exports.
Case studies
Proof-of-work writeups.
Modus Memory: local workstate timelines for agents
The substrate already supports organizer scans, safe moves, OCR, memory search, screen events, and workstate ingestion paths.