AgentfleetAgent workflow orchestration

Compose AI agents into auditable workflows — every step runs with delegated, least-privilege identity.

Workflows, not scripts

Define reusable steps bound to real tools, compose them with conditions, loops, parallel groups and sub-workflows, and publish immutable versions. Move them between environments as packages.

🔐

Delegated identity per step

Each step exchanges the initiating user's token for a short-lived, audience-restricted agent token. Every API call is decided by a policy decision point, tenant by tenant.

🧰

MCP-native tooling

Register any MCP server as a resource and import its tools with one click. Scopes flow into the token exchange, and risky tools can wait for a person's confirmation.

💬

Conversations

Assistant profiles let people chat with an agent that calls tools and starts runs on their behalf — in the console or embedded in a portal as a widget.

⏱️

Automate what runs

Fire workflows on a cron schedule, a Service Bus event or a webhook. Automated runs carry the same delegated identity — no privileged shortcut.

✅

Evals and experiments

Score a workflow with heuristic, LLM-judge and human graders, compare models and prompts side by side, find the cheapest model that passes, and gate promotion to auto-apply.

📊

Built-in observability

Run summaries with tokens, cost and latency, per-step tool and model call logs, a delegation chain, dependency graphs, and a full activity audit trail.

🧮

Usage governance

Budgets and limits on model spend per tenant, workflow, model, user or assistant profile, chat retention, and sensitivity labels that decide which data each model may receive.