coresource

The platform for long horizon agents

Meet sagas, the agent runtime for the longest horizons
SCROLL
The Product

One managed platform for running long horizon agents.

You define the objective. Coresource handles the rest: execution, governance, observability, tools and self-improvement, out of the box. Live today on coresource.ai and through the SDK.

What long horizon means

Long horizon work is measured in days and weeks, not minutes: migrating a hundred thousand records, auditing a codebase for compliance, researching hundreds of gigabytes of private data. Today’s agent platforms assume runs that last minutes. Push them past an hour and they fall apart.

Context collapses
Coherence degrades long before the objective is met.
Failures are fatal
One bad step at hour six kills the entire run.
No controls
Governance, observability and evals were never designed for multi-day work.

At the core of our platform are sagas: a long horizon agent runtime.

The runtime plans with AlphaZero-style search over a learned world model. A proprietary form of Monte Carlo tree search decomposes the objective, simulates paths through a learned world model, and prunes what doesn’t matter before committing.

objective
simulated branches
pruned paths
committed path
Longest Saga
50 hrs
One continuous run · no human in the loop
1.5B
Tokens of context · coherent, hour after hour
50K
Lines of working code · 113 commits, 120 files
201 self-directed plan revisions mid-flight
8 milestones, 36 steps, 8 execution waves
Verified itself to completion against the objective

A single executor run implementing steerability across a live codebase. Tokens processed, not generated: ~1.37B cache-read context, ~4.3M newly written.

Start your first sagaSee the benchmarks