Triggers
Pull requests, issues, pushes, schedules, and the mobile client.
Agentic Workflow · Architecture
Under the hood
One broker, governed execution, controlled outputs.
Every run follows the same path through the system. A workflow is designed in the browser and published to the catalog. The broker, the service at the core of the control plane, governs its execution, and agents do the work on your infrastructure.
01 · System map
From trigger to delivery
-
-
Catalog and policy
Workflow templates, rules, and agent definitions.
-
Dispatcher
Validation, compilation, and task routing.
-
Runtime core
The workflow engine, run state, and human gates.
-
Agents
Workers running on your own infrastructure.
-
Connectors
GitHub, Microsoft Teams, and local environments.
Cross-cutting services, present at every stage:
- Agent registry
- Telemetry and FinOps
- Audit and cost
- Designer and Manager
02 · Providers
One broker, many providers
The agent registry connects the model providers your organization already uses to one governed core. Mix providers per step of a workflow. No vendor lock-in.
- Claude
- Codex
- OpenAI
- Azure AI Foundry
- AWS Bedrock
- Ollama (local)
- Custom APIs
03 · Guarantees
What the design guarantees
- Durable state. Run state lives in the broker. Runs rehydrate after a restart, including runs waiting at human gates and fan-out barriers.
- Policy before execution. Each step passes policy checks before it runs: permissions, owner approvals, and an egress filter.
- Secrets stay local. Credentials live in the native system store, never in a repository.
- Controlled outputs. Every result leaves through a controlled channel: a pull request, an issue, or a report. Nothing is written straight to main.
04 · Demo
See the whole loop on your own repository
A 30-minute demo on a real workflow, not a slide deck. Write to hello@benchframe.com and we will set it up.