REFLEX LABS / EARLY ACCESS
Reusable workflow memoryfor AI agents.
Reflex helps developers and engineering teams using Claude Code and other AI agents capture execution evidence and reuse relevant saved workflows across tasks.
RECORDED PRODUCT DEMONSTRATION
See a saved workflow in use.
In this recorded task, an agent checks synthetic travel expenses against budgets and produces a verified report. The comparison shows the same model working with and without a retained Reflex procedure.
- Agent
- 28,125 tokens
- Agent + Reflex
- 5,415 tokens
- Result
- Both reports verified
One recorded comparison using gpt-6-astra, with synthetic inputs. This is not a Claude benchmark or a promise of savings. Original preparation, final validation time, and infrastructure costs are excluded; Reflex progress is illustrative. Read the source measurements.
WORKS WITH CLAUDE CODE
Keep the method.
Give the agent a head start.
Reflex is a software product for developers and engineering teams who repeat useful work with AI agents. Its Claude Code integration uses native hooks and Model Context Protocol (MCP) tools to connect the agent to saved workflows.
Capture execution evidence
In projects with learning enabled, the local integration records minimized task and tool evidence for Reflex’s learning service.
Retrieve relevant workflows
Reflex looks for saved procedures that may apply to the next task and makes them available to the agent.
Keep Claude in control
Claude reasons about the task, uses its tools, and checks the result. Retrieval alone does not prove that a workflow is suitable or that it saved tokens.
Available in early access
We are developing and evaluating the local client, Claude Code and Codex integrations, and hosted workflow service. The broader vision is shared learning across an organization’s AI work. We measure correctness, token usage, and execution time as the product develops.
Talk to us about your workflows →Our direction: connect AI systems, coordinate their work, and refine execution.
The system we’re building
Connect
Bring models, agents, and tools into a shared view of the work.
Coordinate
Match work to the capabilities it needs, within the company’s rules.
Refine
Use execution results to improve cost, speed, and reliability.
THE THESIS
An organization’s AI should work as one system.
A company’s AI work spans models, agents, and tools. How those parts work together determines the cost, speed, and quality of the result.
Our goal is a shared layer for directing that work—one that uses execution evidence to improve how the system runs, while keeping the organization in control.
IIIAn invitation
Build on what
your agents learn.
Using Claude Code or other AI agents in your team? Tell us which workflows you repeat and help shape Reflex during early access.