ElasticClaw is a runtime for building software workflows.
A programmable, local-first orchestration layer for concurrent coding agents, execution graphs, delegation, plugins, retries, and autonomous development workflows.
elasticclaw-config.yaml + workflows/ci-repair.yaml
schema_version: v1
name: ci-repair
repositories: ["elasticclaw/*"]
env:
NODE_ENV: test
CI: "true"
provider: exedev
---
name: github-issues
trigger:
github_issues:
event: issue_labeled
repositories: ["elasticclaw/*"]
states: [open]
labels: [ci-failing]
concurrency_group: ci-repair
secret_refs:
GITHUB_TOKEN: scoped_github_app
enable_manual_trigger: trueRuntime, not UI
Not an assistant. A runtime.
ElasticClaw is not another place to type prompts. It is infrastructure for systems where agents are scheduled, given scoped credentials, routed through execution graphs, and kept alive across the real lifecycle of engineering work.
Use it to build your own lights-out development systems: workflows that coordinate workers, delegate work, retry failures, react to external events, and preserve state between issue, code, CI, review, and merge.
Core primitives
Compose autonomous workflows from workers, graphs, plugins, retries, and state.
Concurrent agents
Run many isolated agents from the same ElasticClaw Server with workflow-level concurrency groups, queueing, and lifecycle state.
Execution graphs
Model work as workflow stages with triggers, transitions, on-enter actions, terminal states, and issue/PR events.
Delegation and reentrancy
Workflows can start work from external events, manual inputs, or stage transitions, then re-enter the workflow when CI, review, or issue state changes.
Plugin-based extensions
Attach MCP servers, environment variables, model providers, issue trackers, sandbox providers, and workspace files to each class of work.
Local-first operation
Run ElasticClaw Server yourself. Keep configuration, workspaces, workflows, credentials, and workflow state under your control.
Retries and recovery
Feed CI failures and review feedback back into running agents, retry provider bootstrap paths, and terminate cleanly at terminal states.
Logs and task state
Inspect agent status, provisioning events, activity logs, failure summaries, and workflow state from the CLI and web UI.
Build your own workflow
Build the workflow. Don't rent the assistant.
ElasticClaw gives you the runtime primitives to build your own software workflow. Define the workstream, choose the provider, wire in tools and credentials, then let the runtime carry work across the systems your team already uses.
Linear ticket to PR
Move an issue into a trigger state, provision a workspace, inject issue context, open a PR, and advance the tracker when the work is done.
CI repair swarm
Route failed checks back into one or more running agents so the system can attempt targeted repairs instead of dropping state.
PR review workflow
Keep an agent alive after the first PR so it can react to review comments, changed checks, and merge or close events.
Dependency upgrade workflow
Create repeatable upgrade lanes that pull context from releases or issue queues, test changes, and drive PR lifecycle.
Research -> plan -> code -> test
Compose longer workflows from staged instructions, workspaces, tools, environment, and provider-backed execution environments.
$ brew tap elasticclaw/elasticclaw && brew install elasticclaw
$ elasticclaw hub init
$ elasticclaw workspace create --name ci-repair
$ elasticclaw workspace push ci-repair
GitHub-first OSS
Open infrastructure for teams that want to inspect, extend, and operate their own agent runtime.
ElasticClaw is Apache 2.0 and built in the open. Read the code, adapt the workflows, connect your own tools, and run the control plane where you need it.