Not a harness
Use Codex, Claude Code, OpenClaw, Pi, or any shell-capable harness.
Built for agents, not another agent harness
Dispatch complex work.
Keep the thread.
`ldgr` gives agents durable work state for long-horizon tasks, complex multi-step builds, clean interruption, and instant context hydration.
4retained ProgramBench tasks completed
Inspect the evidence →
Inspect the evidence →
01dispatchcomplex build
02observeevidence recorded
03interruptcontext cleared
04hydrateldgr status
05resumework continues
fresh contextfull thread
Start with core
One CLI. No new control plane.
curl -fsSL https://raw.githubusercontent.com/hydra-dynamix/ldgr-core/main/scripts/install.sh | shA real `ldgr-explore` demonstration
Give it a repository.
Get back the thing you missed.
Watch an agent investigate a public repository, lose its context halfway through, hydrate from `ldgr`, and continue to an evidence-backed finding.
Preview the demoThe demo uses a paid adapter. Core remains free and open source.finding 01evidence-backed
The public API and the runtime dependency graph disagree.
entrypointdispatchhidden branch
3 artifacts8 source references0 guesses
Purposefully narrow
It is not where you
babysit an agent.
Not a dashboard-first control plane
Dispatch commands through your harness and let the agent work.
Not an MCP or tool catalog
The self-contained CLI is the interface. A shell is enough.
Not another bag of skills
Skills only teach a harness how to operate the durable substrate.
Open source
The durable substrate.
Open source
ldgr
The durable work ledger agents use to continue complex tasks across interruptions, context resets, and handoffs.
Explore product →Open source · alphaldgr-research
A research discipline for agents: programs, hypotheses, experiments, facts, evidence, and readiness gates.
Explore product →Open sourceldgr-programbench
The inspectable proof adapter for four retained LDGR-assisted ProgramBench reproductions.
Explore product →Commercial adapters
Focused systems for harder work.
The next context can know what the last one learned.