INVESTMENT // 2026
Every major AI system today can generate. None can be trusted to finish a long task alone. The gap between output and completion is reliability: the planning, recovery, and verification that let an agent commit across hours of unsupervised work and actually deliver. We are building that layer, and the proof that it works.
02 / THE THESIS
The dominant frame for AI progress is capability: larger models, faster inference, more training data. Capability is necessary but not sufficient. A system with world-class reasoning still has to hold a long task together across hundreds of steps without a human watching. Pointed at a long horizon, today's agents drift, wedge, hallucinate progress, and quietly fail. The model is not the problem. The agent around it is.
Emotions Research treats long-horizon reliability as an engineering discipline, not a prompt trick. We do not train models. We build the agent layer on top of any model you bring: a planning and self-supervision loop, recovery from stalls and wedges, and a verification gate that lets a task land only when it actually passes. The result is not a smarter model. It is an agent that finishes the work and leaves an auditable trace of how.
03 / WHY NOW
CAPABILITY // PRESENT
Foundation models can reason, plan, and execute. They cannot self-supervise over long horizons without drifting. The capability is there. The autonomy loop is missing. That is an architecture problem, not a model problem.
COST // FALLING
Driving an agent across hundreds of steps with full verification was cost-prohibitive 18 months ago. Token and inference costs have dropped such that long-horizon autonomous runs, the kind that finish real work, are now economically viable for production systems at scale.
MARKET // NOW
Enterprise AI adoption stalls at the boundary of unsupervised action. Customers want agents that work through the night without babysitting. The limiting factor is not model quality. It is trust. A reliable long-horizon agent is the unlock.
04 / WHAT WE ARE BUILDING
Workstream 01
The planning and self-supervision engine that keeps an agent on a goal across hours of work: task decomposition, progress tracking, and recovery when a step stalls or wedges. Built for runs with no human in the loop and no fallback to "ask the user."
Workstream 02
Every change runs in an isolated git worktree and is gated on types and tests before it lands. A task is done only when it passes, with an auditable trace of what ran. Ship on green, not on hope. This is what makes unsupervised work safe to trust.
Workstream 03
The instruments that keep a long run alive: heartbeated brain loops, a supervisor that restarts a wedged agent, and structured fleet telemetry compatible with existing model providers. You cannot trust what you cannot see. We build the reliability layer first.
Workstream 04
Our first deployed system. A local-first coding agent that dispatches parallel worker fleets, gates output on types and tests, and ships reviewable branches without human steering. Nyx is research made concrete at production scale. See it run in the Terminal.
05 / THE PATH
A pre-seed is a milestone purchase, not a runway. This round buys 24 months and a single deliverable: a working long-horizon autonomy loop with published reliability benchmarks underneath it, and paying teams on Nyx as the proof. The allocation and the gates below are the plan we are accountable to.
THE 24-MONTH GATE PLAN
MONTH 0–6 // FOUNDATION
Public OSS release of the Nyx fleet, the autonomy loop hardened against wedging, and the first reliability harness measuring completion, recovery, and drift across real worker runs. Gate: instrumented runs in production, not a notebook.
MONTH 6–12 // BENCHMARK
The first quantitative benchmark for whether agents finish long unsupervised tasks, with the reliability dashboard in private beta. Gate: a reproducible benchmark others can cite and a design partner running Nyx against their own work.
MONTH 12–18 // TRACTION
Paying teams on Nyx with a reliable autonomy loop driving unsupervised long-horizon work. Gate: recurring revenue, OSS adoption at scale, and retention that proves the autonomy is trusted, not babysat.
MONTH 18–24 // SEED
A proven autonomy architecture, published reliability benchmarks, and a revenue line. Gate: a seed round priced on traction and results already shipped, with this pre-seed having de-risked the thesis end to end.
06 / THE ASK
PRE-SEED ROUND
The capital funds a core team of four, 24 months of long-horizon agent runs, and the evaluation infrastructure to establish reliability benchmarks that do not yet exist. We do not train models. We are not raising on a demo. Nyx is deployed and being used today. We are raising to harden the autonomy loop underneath it and publish the first quantitative benchmarks for whether agents finish real unsupervised work.
We are not building toward an exit in five years. We are building the agent layer for the next decade of AI. The right investors are those who understand that reliability research and product traction are the same bet at this stage.