Oracle announced Fusion Claw on September 29, 2026. Oracle calls it a governed agentic execution runtime for Oracle Fusion Agentic Applications. A frontier model reasons and plans, then deterministic computation does the calculations, reconciliations, and simulations at scale. It shipped with 25 new Claw-powered agentic applications across finance, HR, supply chain, and sales, added to a portfolio Oracle puts at 75 Agentic Applications.
The name invites a comparison with OpenClaw. TechTarget reports that Oracle senior vice president Natalia Rachelson said Fusion Claw was inspired by OpenClaw but is aimed at enterprise work. The two do very different jobs. Fusion Claw runs inside Oracle's Fusion Cloud Applications against your system of record. OpenClaw and Hermes are agent runtimes you install and run yourself.
Sources for the Oracle side are the Oracle announcement (September 29, 2026), Oracle's Fusion Insider post (October 13, 2026), the Fusion Claw capabilities page, the Configure Autonomous Outcomes guide, and TechTarget's launch coverage, all checked on October 16, 2026. The OpenClaw and Hermes side reuses our OpenClaw vs Hermes Agent comparison.
The short answer
Look at Fusion Claw if your finance, HR, supply chain, or sales data already lives in Oracle Fusion Cloud Applications and the work is long, data-heavy, and rule-bound: closing the books, staffing schedules, load planning, sourcing awards, territory plans. Oracle supplies the runtime, the governance layer, and the audit record. You set the objectives, rules, and how much autonomy each application gets.
Use OpenClaw or Hermes when the agent has to work across tools you pick, answer in chat or messaging channels, run on a schedule on a host you control, or serve a business that does not run on Fusion. Start from trust boundaries for self-hosted agents and permissions and human approval.
What Fusion Claw is
Oracle splits each run into two parts. The model reads the objective, analyzes the problem, weighs options, and writes a plan. Deterministic computation then runs the repetitive, high-volume steps. Oracle says this can lower cost because AI is used only where reasoning is needed. Once a run starts, Fusion Claw can revise its plan when conditions change without the user staying involved at each step, and it keeps context from completed work so similar tasks can build on earlier results.
Oracle says Fusion Claw runs on Oracle Cloud Infrastructure and uses frontier models including Gemini and OpenAI, with more models planned. TechTarget describes the runtime as software that holds the OS and other code the models need to carry out tasks.
The three governance pieces
Oracle names three parts:
- Enterprise Operating Envelope. The authority layer. It holds the organization's objectives, standard operating procedures, policies, constraints, permissions, risk thresholds, decision rights, approval requirements, and escalation boundaries.
- Outcome Trust Harness. Enforces the envelope on each run. Oracle says it governs identity, which data and capabilities Fusion Claw can use, which actions it can take, and when approval or escalation is required.
- Outcome Receipt. The audit record after a run finishes: authority applied, evidence used, decisions made, actions and transactions executed, and the result.
Oracle also describes two boundaries. The reasoning layer is kept apart from the authority and execution layers, so the model cannot grant itself more permissions. Deterministic execution runs in a secure, temporary environment, so the model cannot reach or change the underlying data and system of record directly. One analyst quoted by TechTarget, Mark Vigoroso of The Enterprise Edge, said tasks will run in isolation without open access to the internet or APIs. That comes from his briefing. We did not find it in Oracle's own materials.
Autonomy levels
Each agentic application gets its own autonomy setting. Oracle's Fusion Insider post lists three: require approval before actions, let Fusion Claw work while people direct or step in, or allow fully autonomous execution within set permissions. The press release frames the range as quick assistance up to governed full auto inside explicitly delegated authority.
The configuration guide shows how an admin sets this up in AI Agent Studio. You turn on autonomous outcomes for an application, then define goals written as results, skills (task instructions with decision logic, validation, evidence rules, blocker handling, and completion conditions), extra workflow agents, and data extracts that call a PL/SQL stored procedure. Oracle says data security is enforced in the database before data comes back, so an extract returns only what the user may see. Examples are tagged Research, for tasks that do not change business data, or Full Auto, for tasks that complete work and take actions. The guide also lets you upload a skill file built with tools such as Codex.
What it ships with
Oracle lists the 25 Claw-powered applications as part of Release 26D. Examples named across Oracle's pages include:
- Ledger, for close readiness: account variances, usage against controls, and anomalies before close.
- Collector Workspace, for prioritizing collections and running policy-permitted collection actions.
- Workforce Operations Command Center, for schedule balance simulations that keep hard scheduling constraints.
- Shipment Load Planning Workspace, for rough-cut load plans and consolidation options.
- Sourcing Command Center, for what-if supplier award scenarios against your award policies.
- Account Territory Growth Plan, for account-level whitespace and revenue risk.
- A CPQ test-to-production migration agent that simulates a configuration package against production data before it goes live.
On pricing and timing, TechTarget reports Rachelson said the apps arrive in the fourth quarter and are free with Agentic Applications subscriptions, which are priced separately from Fusion Cloud Applications, with usage tied to AI units. We have not seen that in Oracle's own pages, so confirm it with Oracle.
OpenClaw and Hermes run a different kind of job
OpenClaw is a Gateway you run. It connects model providers, sessions, tools, messaging channels, the Control UI, and optional device nodes. It extends behavior with skills, plugins, and MCP servers. Sandboxing is off by default. When you turn it on, tool execution can move to Docker, Podman, SSH, OpenShell, or Crabbox while the Gateway stays on the host.
Hermes Agent is a process you run from a CLI, a messaging Gateway, an API server, or a supported editor. Tools come in toolsets you select, skills load on demand, and it can connect to local or remote MCP servers with per-server tool filtering. Dangerous commands go through an approvals setting (`smart`, `manual`, or `off`) with a hardline blocklist that stays on underneath.
Some of the ideas map across. Fusion Claw's envelope is close to the tool policy, deny rules, and approval settings you write for OpenClaw or Hermes. Its harness is close to the Gateway and sandbox that enforce them. Its receipt is the audit trail you have to build yourself on a self-hosted stack. With Fusion Claw, Oracle owns the runtime and the system of record. With OpenClaw or Hermes, you own the host, the tools, and the logs.
Side by side
| Question | Oracle Fusion Claw | OpenClaw or Hermes |
|---|---|---|
| Where does it run? | Inside Oracle Fusion Cloud Applications on Oracle Cloud Infrastructure | A host and sandboxes you configure |
| What job is it for? | Long-running finance, HR, supply chain, and sales work on Fusion data | Chat, schedules, tools, and channels you wire yourself |
| Who starts the work? | A user defines an outcome in natural language inside an agentic application | Chat messages, schedules, API calls, or a person |
| How are actions gated? | Enterprise Operating Envelope enforced by the Outcome Trust Harness; per-app autonomy from approval-first to full auto | Tool policy, toolsets, deny rules, sandboxing, and approval settings in config |
| Audit record | Outcome Receipt per run | Logs and records you set up |
| Data access | Fusion capabilities and stored-procedure extracts with database-level data security | Whatever credentials and tools you grant |
| Models | Frontier models including Gemini and OpenAI, per Oracle | Provider choices you configure |
| Who maintains it? | Oracle | You install, patch, and audit the runtime |
Distinct from other managed agent runtimes
- OpenAI Agents API: a hosted agent harness you call by API with MCP and a hosted or self-hosted sandbox. It is general purpose. Fusion Claw is a runtime for Oracle Fusion Agentic Applications.
- AWS AgentCore Runtime V2: managed compute for agents you build and deploy. Fusion Claw comes with Oracle-built applications and Oracle's governance layer.
- Finance approval gates: our guide to where a person should approve money-moving steps. The same questions apply when you pick a Fusion Claw autonomy level.
Before you turn on full auto
- 1Start each application on approval-first or Research mode and read the Outcome Receipts before raising autonomy.
- 2Write the envelope as you would a policy for a new hire: which accounts, amounts, suppliers, or employees are in scope, and what must go to a person.
- 3Check every data extract with your DBA. Oracle says database security limits what comes back, so the user role behind the run matters.
- 4Treat skill files as code. Review them before upload, since they carry the decision logic and completion rules.
- 5Confirm availability, pricing, and AI unit usage for your Fusion subscription with Oracle.
- 6Use action budgets thinking for money-moving apps such as Ledger or Collector Workspace: cap amounts and categories before full auto.
How this fits implementation work
OrchestriAI does not resell or operate Oracle Fusion, OpenClaw, or Hermes for clients. AI agent systems decide which jobs an agent may run and where a person approves. Systems integration connects agents to the systems that hold the data, including ERP. Custom development covers the internal tools around them.
Independence and limitations
OrchestriAI is an independent implementation provider. Product names are used for identification only. OrchestriAI is not affiliated with, endorsed by, or sponsored by Oracle, OpenClaw, or Nous Research / Hermes. This article summarizes Oracle's announcement (September 29, 2026), Oracle's Fusion Insider post (October 13, 2026), the Fusion Claw capabilities page and Configure Autonomous Outcomes guide, and TechTarget's coverage, all checked on October 16, 2026, plus our existing OpenClaw and Hermes guides. We have not run Fusion Claw ourselves. We do not claim MCP support, a specific model list beyond Oracle's statement, pricing beyond what TechTarget reports, or that Fusion Claw is related to the OpenClaw project in code. Features and terms can change, so confirm current Oracle documentation before you grant an application write access. This is not a security certification or a compliance conclusion.
