Building and Deploying Agents on Reindeer

Reindeer provides a secure runtime and builder harness for enterprise agents. Teams bring the agent logic, workflow policy, prompts, and tools they want to run. Reindeer provides the runtime services around that agent: identity, secrets, auditability, evaluations, human review, observability, and enterprise integrations.

The result is a place to build, run, and improve agents without rebuilding the operational controls that enterprises need before production deployment.

Quick Start

If you are using an agent harness, code assistant, or internal builder workflow, point it at the Reindeer skill URL:

https://storage.googleapis.com/reindeer-release-external/skill/SKILL.md

The skill file is the source of truth for programmatic use. It tells the harness how to install the CLI, authenticate, create and update resources, trigger runs, and inspect results.

REINDEER_SKILL_URL="https://storage.googleapis.com/reindeer-release-external/skill/SKILL.md"

# Configure your harness or agent builder to read this URL as its Reindeer skill reference.
# Then use the documented CLI commands for workspaces, agents, artifacts, contexts, connectors, and runs.

For non-interactive harnesses, CI jobs, and scripts, use headless authentication:

reindeer login --headless --no-prompt

Where Reindeer Fits

Reindeer sits alongside an existing agent stack. It does not require teams to migrate source systems, replace orchestration frameworks, or move operational data into a new system. Agents can be defined and managed through the Reindeer CLI, API, or existing builder workflows, then executed against connected enterprise systems through scoped integrations.

• Bring your own agent logic. Use Reindeer to package prompts, policies, tools, contexts, and workflow behavior into a governed agent.
• Use Reindeer as the execution harness. Trigger runs from connectors, CI workflows, scheduled jobs, or external systems.
• Keep enterprise controls in the runtime. Secrets, identity, permissions, audit logs, feedback, and escalation are handled by the platform.

Security and Runtime Controls

• Private cloud and dedicated environment options. Reindeer can run in a dedicated environment so customer data remains isolated according to enterprise requirements.
• Enterprise identity support. Teams can connect their own identity provider and SSO so users are provisioned through existing identity management processes.
• Per-workflow secrets and scoped access. Credentials are scoped to the workflow and tool that needs them. Secrets are injected at runtime and are not exposed directly to the agent.
• Compliance posture. Reindeer is designed for enterprise governance requirements including SOC 2 Type 2, ISO 27001, and GDPR-aligned controls.
• Full audit trail. Tool calls, decisions, escalations, inputs, outputs, and human interventions are logged for traceability.

Quality Controls for Production Agents

• Confidence scoring. Each action can be evaluated against workflow-specific thresholds before an agent proceeds automatically.
• Human-in-the-loop escalation. When confidence is low, policy requires review, or an exception is detected, the case is routed to a human with the relevant context and reasoning trace.
• Plain-English feedback. Operators can correct behavior in natural language. Reindeer uses that feedback to update agent logic and rerun affected cases where appropriate.
• Pattern detection. Repeating exceptions are surfaced as proposed fixes that can be reviewed and approved instead of becoming a long list of one-off tasks.
• Run history and replay. Historical runs can be inspected and replayed to validate behavior after logic, context, or tool changes.

Pre-Built Skills and Enterprise Integrations

Reindeer includes hundreds of pre-built skills and integrations that builders can attach to agents instead of implementing common enterprise capabilities from scratch. Examples include OCR, document parsing, information extraction, classification, routing, and connectors for systems such as SAP, Oracle, NetSuite, Workday, ServiceNow, and other enterprise applications.

Skills are invoked through the runtime with scoped permissions. This keeps connector behavior, credential handling, logging, and error reporting consistent across agents.

Install the CLI

Install the Reindeer CLI on macOS or Linux:

curl -fsSL https://storage.googleapis.com/reindeer-release-external/install.sh | bash

Install the Reindeer CLI on Windows:

irm https://storage.googleapis.com/reindeer-release-external/install.ps1 | iex

Authenticate

Use interactive login for local development:

reindeer login

Use headless login for CI, scripts, automation harnesses, or non-interactive environments:

reindeer login --headless --no-prompt

Build and Run a First Agent

A typical CLI-based builder flow creates a workspace, registers an agent, attaches knowledge, triggers the workflow, and inspects runs:

reindeer workspaces create --name my-workspace
reindeer agents create     --file agent-config.json
reindeer artifacts create  --file knowledge-doc.pdf
reindeer connectors trigger --file run-config.json
reindeer runs list

Core Concepts

Harness and CI Integration

Reindeer can be used as a harness around existing agent development workflows. Builders can define agents and contexts declaratively, trigger runs from CI or scripts, and inspect outcomes through the CLI or API. This supports repeatable validation before promotion and allows teams to compare behavior across versions.

• Use --profile to separate development, staging, and production environments.
• Use -w or workspace-specific configuration to target the correct workspace.
• Use -o to produce machine-readable output for CI pipelines and automated checks.
• Use headless login for scheduled jobs, test harnesses, and deployment automation.
• Use run history and replay to inspect regressions after changes to prompts, context, tools, or policy.

Architecture Notes

Reindeer runs on top of the enterprise stack already in place. It integrates with source systems through connectors, executes agent workflows with scoped tools and credentials, and records run-level observability and audit data. The platform is designed to respect existing data governance, access controls, and operational boundaries.

Builder Tips

• Start in a development workspace before promoting workflows to production.
• Tune confidence thresholds per workflow. Defaults are a starting point, not a final operating policy.
• Version contexts so prompts and reference documents roll out in a controlled way.
• Watch runs and cases first when diagnosing quality issues; this is where exceptions and regressions surface earliest.
• Keep tool scopes narrow. Agents should only receive the connectors and credentials required for the workflow they execute.

CLI Reference

The full CLI reference is available at Reindeer CLI docs. It covers installation, authentication, resources such as agents, connectors, artifacts, and contexts, workflow execution, and multi-profile setup.