The automation agent for teams that need their AI workflows to learn as they run.
Dobby is the workflow execution layer in the LocusGraph ecosystem. It is what you use when you need agents executing tasks autonomously — writing code, running pipelines, processing work queues — and you need those agents to get smarter the longer they operate, not reset every time.
Running an AI agent on a task is easy. Running an AI agent on a class of tasks — reliably, over weeks, with minimal correction — is where things break down.
The problem is not the model. Models are capable. The problem is that workflow agents operate without memory of what they have learned. Every edge case they navigate, every shortcut that backfired, every pattern that saved time — none of it carries forward. The next run starts fresh.
Dobby is built to fix that specific failure mode.
Dobby runs your automations and emits structured events to LocusGraph as it works. Every decision, correction, and outcome becomes an input to LocusGraph's admission pipeline. What comes back is knowledge — typed, linked, confidence-scored — that Dobby can query on the next run.
The loop looks like this:
- Dobby executes a workflow task
- Significant events — errors resolved, decisions made, patterns confirmed — get admitted to LocusGraph
- LocusGraph processes those events into structured knowledge
- On the next run, Dobby queries that knowledge before acting
Over time, the workflow does not just run. It improves. It knows which approaches have worked, which have failed, and how confident to be in its own outputs.
- Coding workflows. Agents that write, review, refactor, and debug across sessions, carrying forward what they have learned about your codebase's patterns and constraints.
- Support automations. Agents that handle incoming requests and accumulate structured knowledge about resolution patterns, edge cases, and escalation triggers.
- Data and ops pipelines. Workflows that execute against live data and build inspectable knowledge about what the data looks like, how it changes, and what breaks things.
Dobby connects to LocusGraph's knowledge graph, not to any single LLM. Swap your underlying model and the knowledge your workflows have accumulated stays intact. It belongs to your team.
Dobby is built on top of LocusGraph's structured knowledge infrastructure. LocusGraph handles the admission and graduation pipeline — turning raw agent events into typed, temporal knowledge. Dobby handles the execution layer that generates those events and queries the knowledge that results.
They are designed to be used together. Start with Dobby if you are running automations. Start with LocusGraph if you are building a custom agent and need the knowledge layer directly.
See the LocusGraph documentation at locusgraph.com for full setup, integration patterns, and how Dobby connects to the knowledge graph.