The Orchestrator

OpenHuman is an orchestrator, not a chatbot: durable agent graphs, visual workflows, sub-agent fleets and an always-on two-agent layer.

Most harnesses run one agent in one loop. OpenHuman coordinates many agents over long stretches of time, across machines. It does this durably, visibly and under your control. Four layers make that work.

1. Graphs, not loops

Every agent turn runs on tinyagents, our open-source graph engine. Multi-step work compiles to state-machine graphs with conditional routing:

plan → execute ⇄ review → finalize for delegation.
Phase DAGs for multi-agent workflow runs.
Map-reduce fan-out for parallel workers.

All of it is checkpointed. A graph can pause mid-run, for your answer, an approval or a restart, and resume exactly where it stopped.

2. Sub-agent fleets that do not get lost

The orchestrator spawns specialized sub-agents, up to 3 levels deep. It reuses compatible idle workers instead of spawning new ones. It also routes each worker to the right model tier: heavy reasoning for the core, and a fast burst tier for low-context workers.

Reliability is built in. A no-progress circuit breaker stops loops. A stuck child hands back a question (pause and resume on your answer) or an Incomplete root-cause summary, never silence. See the agent harness.

3. Workflows you can see

Workflows move orchestration out of the chat. The agent proposes a typed graph of triggers, agents, tools and conditions. You review it on a canvas and save it. Runs are durable, gated by approvals and inspectable step by step. They run on the open-source tinyflows engine.

4. An always-on two-agent layer

Inbound traffic first reaches a fast reflex agent that triages in seconds. It hands a concise brief to a deeper reasoning core, which does the multi-step work and delegates to workers. You can steer a task mid-run, because input can be delivered into a live session. 20:1 compression keeps week-long sessions bounded.

What is next

Agents writing control flow as small programs in a sandboxed REPL is not being built. The harness dropped that runtime and the tool no longer exists. Graphs with checkpointing under a trust model are the shape orchestration takes.

Two items are on the roadmap: an agent-to-agent protocol so a graph can span instances, and durable swarm task graphs.

How it differs

Single-agent harnesses (Claude Code, OpenClaw, Hermes)

OpenHuman

Execution model

One loop, one context

Compiled graphs, conditional routing, checkpoint and resume

Parallelism

Manual or plugin

Native sub-agent fleets, map-reduce fan-out, worker reuse

Automation

Scripts and cron

Visual, durable, approval-gated workflows

Always-on

None

Reflex agent plus reasoning core, mid-task steering