For Enterprise

Scale AI agents
across your team without the chaos

Your developers are already using AI coding agents. Ocean gives you the platform to make that productive, auditable, and safe — without slowing anyone down.

What you can't see can't be managed

Every developer on your team is running AI agents differently. Different terminals, different workflows, no shared context. Some use Claude Code, others use Cursor or Codex. None of their tools talk to each other.

When two agents edit the same file, someone loses work. You have no visibility into what agents are doing, how many tokens they're burning, or what code AI actually wrote. Shipping AI-generated code means trusting a process you can't see or audit.

You need to enable AI-assisted development — it's a competitive advantage — but you also need governance, cost control, and a clear audit trail.

A managed platform, not ad-hoc experimentation

Ocean turns AI-agent development into a managed capability. Every agent session is a node in a persistent DAG — every unit of work has a stable identity that an audit entry can point at. Copy-on-write isolation prevents agents from corrupting shared state.

A four-tier managed configuration hierarchy gives IT a real policy surface: pin models, restrict endpoints, lock down plugins, and seed workspace defaults at the system level without developers being able to silently override. The immutable audit log and session-level replay turn "what did that agent do?" from an interview into a timestamped record.

All data stays on the developer's machine by default. No terminal content, commands, or session data leaves the device. Telemetry is opt-in with PII redaction.

Governance that enables, not restricts

Every feature is designed to give teams more power while giving managers more visibility.

Session DAG

Every agent session is a node in a persistent directed acyclic graph. Every unit of agent work has a stable identity — which means every audit entry, every forensic question, and every policy decision can point at a specific node instead of a fuzzy "the terminal".

  • Persistent parent/child links across restarts
  • Fork, branch, and subtree rollback as first-class ops
  • Visualizer with Cmd+Shift+G (per-user, not shared)
  • Stable session IDs referenced by audit entries

Audit Log

Immutable record of every terminal operation, file change, git action, and configuration update. Filter by session, workspace, or event category. Export to JSONL for compliance pipelines.

  • Session, workspace, filesystem, git, and config event categories
  • Full metadata logging with timestamps
  • JSONL export for SIEM and compliance tools
  • Paginated viewer with filtering

Session Recording & Replay

Per-session capture of terminal output, commands, and agent activity. When an incident lands, forensics is no longer an interview with the developer — it's a timestamped replay of the exact session, scrubbable and exportable.

  • Full terminal + agent-event capture per session
  • Variable-speed playback with event jump
  • Transcript export for post-mortems and regulators
  • Local-only storage — no vendor cloud required

Managed Configuration

A four-tier hierarchy — system → user → workspace → session — gives IT a real policy surface. Pin model choice, restrict remote endpoints, gate plugins, and seed workspace defaults at the system level. Developers cannot silently override system-tier values.

  • System-tier precedence over user overrides
  • Model pinning and remote-endpoint restriction
  • Plugin allowlist/denylist via permission declarations
  • Workspace templates for standardized team setup

Workflow Engine as a Policy Gate

TOML-defined multi-step agent pipelines with event triggers (git.push, pr.opened, schedule.daily) and a visual DAG editor. Use it as a compliance gate — run a secret-scan on every push, block merges that fail, log both outcomes to the audit stream.

  • Event-triggered pipelines on git and schedule events
  • Per-step model selection and permission scoping
  • Outcomes recorded in the audit log by default
  • Visual DAG editor for non-command-line reviewers

Usage Quotas & Budget Controls

Track token spend per session with configurable budget limits. Alerts at 50%, 75%, and 100% thresholds. Daily and monthly period tracking. Voice notifications for budget events so developers stay aware without checking dashboards.

  • Per-session cost tracking in real time
  • Configurable budget ceilings (daily, monthly)
  • Graduated alerts at 50%, 75%, 100% thresholds
  • Claude API quota monitoring (5h, 7d windows)

Teams & Relay V1.x

Real-time team synchronization via a relay server you can self-host. Shared workspaces with role-based access control (read, write, admin). WebSocket presence shows who's working where. JWT tokens secured in macOS Keychain.

  • RBAC with read, write, and admin tiers
  • Real-time WebSocket presence with auto-reconnect
  • Shared workspaces across team members
  • Self-hostable relay server for air-gapped environments

SSO via WorkOS V1.x

Enterprise single sign-on through WorkOS AuthKit (Google, GitHub, Microsoft). OAuth flow with browser redirect. Session persistence across app restarts. SAML / SCIM are on the V2 track and prioritized on customer demand.

  • WorkOS AuthKit integration
  • Session restoration and deep-link handling
  • User profile management
  • Backend session persistence in SQLite

Conflict Prevention

Cooperative file locks prevent two agents from editing the same critical file simultaneously. Pre-conflict warnings alert developers before edit paths collide. Prevention is always cheaper than resolution.

  • Cooperative file locking across sessions
  • Pre-conflict warnings when paths overlap
  • Conflict heatmap identifies hotspot files
  • Force-break locks for admin override

Workspace Health Dashboard

Aggregate view of workspace health. Health score from 0-100 based on session activity, conflict severity, and resolution metrics. File modification heatmap. Agent dashboard with per-agent token usage and estimated costs.

  • Workspace health score (0-100)
  • Conflict severity tracking
  • File modification heatmap
  • Per-agent cost estimation in USD

Four-tier configuration, one source of truth

System-tier values win. Developers get full control over the settings IT didn't lock down, and zero ability to override the ones it did.

PRECEDENCE overrides below SYSTEM IT-managed — written by MDM or installer model.allowlist = ["claude-*", "gpt-4-*"] · plugins.allowlist = [...] · telemetry = off USER per developer — preferences that don't touch policy theme = "dark" · keybindings = {...} · default_agent = "claude-code" WORKSPACE per project — team-agreed defaults workflow.on_pr_opened = "review.toml" · budget.monthly = 500 SESSION per run — ephemeral overrides for one agent model = "claude-opus" · permissions = ["read_only"]
IT ships system-tier — devs can't override conflicts resolved at the higher tier every change lands in the audit log

100% local by default

Ocean runs entirely on the developer's machine. No terminal data leaves the device unless the developer opts in.

Data Type Storage Shared Externally?
Sessions & history Local SQLite (~/.ocean/ocean.db) Never
Terminal content In-memory only Never
Settings & preferences Local (localStorage + SQLite) Never
Auth tokens (relay) macOS Keychain (encrypted) Never stored in plaintext

Local-First Architecture

All persistent state lives in a local SQLite database with WAL mode for performance. No cloud dependency for core functionality. Relay sync is optional.

Secure Token Storage

JWT and refresh tokens are stored in the macOS Keychain using the apple-native backend. Keychain access is lazy — tokens are only read when explicitly needed, never at app startup.

Self-Hostable Relay

The relay server URL is configurable. Teams can run their own relay instance in air-gapped or restricted environments. No data flows through Ocean's infrastructure unless you choose it.

Ready to make AI agents a managed capability?

See Ocean in action with your team. We'll walk through audit logging, usage controls, team sync, and how Ocean fits into your existing development workflow.

Beta pilots keep Ocean Platform free forever for the team's current seats — including enterprise pilots started during beta.

Request a Team Demo Download and Try