Workflows
Graphs of triggers, code, agents, controls, and approvals.
Local-first · Durable · macOS, Linux, Windows
Describe what you need in plain language. Anselm creates the functions, agents, and workflows, schedules them, and runs them durably on your own machine. No orchestration code, no infrastructure to manage.


One request, start to finish
Every screenshot on this page comes from the same workspace and the same request: a weekly digest of a GitHub repository, reviewed by a person before it is filed.
Ask for the outcome in chat. Anselm plans the work, creates each building block, and reports what it built with links to every entity and run.


The result is a real workflow: a cron trigger, a Function that calls GitHub, an Agent that writes the digest, an approval step, and a Function that files the document. All versioned.


The Scheduler shows what is running, what is waiting on you, what failed in the last 24 hours, and when the next fire is due.


Approvals, failures, and completions land in a durable ledger. Approve from the notification, the run continues, and the draft in the Library is published.


Product
Chat is where work starts. The other surfaces let you inspect, schedule, supervise, and configure everything the AI built.


Where work starts. Ask, attach, mention, and watch entities appear as the conversation goes.
Explore →

Every Function, Handler, Agent, and Workflow with versions, relations, and a live debug console.
Explore →

Triggers, upcoming runs, and a node timeline for every execution, including the ones that resumed.
Explore →

A durable ledger of approvals, completions, and failures, plus native OS notifications.
Explore →

Documents and skills in a native editor, with outline, properties, and backlinks.
Explore →

Models and keys, MCP servers, storage, network, and shortcuts. Managed by default, BYOK when you want it.
Explore →Durable execution
Workflows that run while you are away need more than retries. Anselm records every node result, pins versions for in-flight runs, pauses for human approval when a step needs it, and picks up exactly where it stopped after a restart.
Completed nodes are reused on recovery. External side effects are not repeated.
Editing an entity never changes a run that is already in flight.
Approval nodes park the run and wait for a decision or a timeout policy.
Every run has a node timeline, inputs, outputs, and the exact versions it used.
Capabilities
Graphs of triggers, code, agents, controls, and approvals.
LLM workers with instructions, tools, skills, memory, and sub-agents.
Sandboxed code, stateless or stateful, created and maintained by AI.
Documents, skills, and memory available to every agent and conversation.
Images, audio, video, and documents flow through chat, agents, and workflows alike.
MCP servers, web search, and your own tools in one curated catalog.
Deployment
Anselm runs as a desktop app with all data in local SQLite and files. Use managed models through the Anselm API with no key setup, or bring your own provider keys.