Today we’re releasing Coral v1: the execution rails for agents.
And we’re sharing a preview of what’s next: Sailfish, our open-weights model for enterprise tool calls.
Agents are moving from answering questions to executing work in the systems that run your business. They read sensitive databases, issue refunds, recover production services and more. But these use cases are bottlenecked by controlling the risk.
Agents also start every task from scratch. They re-read tool schemas, retry failed calls, and spend tokens on steps they got right yesterday.
Coral fixes both these problems.
A tool is the wrong unit of control
Teams usually connect agents to enterprise systems in one of two ways, both with flaws.
- Give the agent raw MCP tools. Permissions are all or nothing per tool. If an agent can call
update_record, it can update any record in any way the tool allows. The agent also works out every multi-step task again each time it runs. - Hand-build an integration for each agent. You get control, but your team now maintains glue code for every agent, system and use case. That code doesn’t keep up with all the agents’ tasks.
How Coral works
Agents write programs. To use tools, an agent writes a short TypeScript program. One program can call several tools, filter the results and return only what the task needs. Less raw data passes through the model’s context.
Coral runs them on rails. Each program runs in a sandbox, system credentials never appear in the agent’s context or code, and every call is checked against your policy before it reaches the system behind it. Policies can also require a person to approve it.
What works becomes a custom tool. When a program succeeds, the agent saves it to Coral as a custom tool. Your library of proven tool-calling code grows with each task. Agents search that library and reuse, edit and share custom tools across your team instead of starting over.
Why functions change the economics
Precise control. You can limit agents to approved functions only, not everything an MCP tool permits. Security approves “look up an order’s refund status,” not the whole payments API.
Cheaper models, same accuracy. With custom tools, a smaller model performs like a bigger one. For many tasks, this means Opus 5.5 accuracy at Haiku 5.5 cost.
Faster, leaner agents. With custom tools, agents complete the same tasks 29% faster and at 31% lower token cost on average. See examples.
Training data you own. Each run leaves a structured trace of the goal, the function used and the outcome that works well for post-training your own model.
Built for enterprise systems
- Your infrastructure. Coral runs anywhere from a laptop to your own cloud, with zero data egress.
- Open source. Coral is Apache 2.0. Coral Enterprise adds organization-wide governance and support.
- Any agent. It works with any AI app or agent that supports MCP.
- Any system. You can connect anything with an MCP server or OpenAPI spec, including more than 5,000 databases and SaaS tools.
Already in production
Thousands of developers at more than 450 companies already use Coral for customer support, SecOps, business intelligence, SRE and hundreds of other use cases.
“Coral gives agents one place to find and call the APIs our business runs on. Our teams no longer wait on scripts or exports for everyday questions, and bringing a new internal system online for agents is trivial. Saved functions guide the agents, so they now make fewer tool calls, use fewer tokens, and return fewer wrong answers.”
— Andrew Howell, VP of Engineering, Blockdaemon
Preview: Sailfish
Sailfish is our new family of open-weights models built for enterprise tool calls. We are using it internally and will share how you can deploy it soon. Get an email when the weights are released.
Agents that answer before you’ve switched tabs. When an agent triages an alert or answers a support ticket, every second counts. Frontier models are smart but slow at the many small calls this work needs. Sailfish is efficient, so agents finish in a fraction of the time, at a fraction of the cost.
Nothing leaves your infra. With Coral your data already stays on your infrastructure. With Sailfish your model stays there too. You run it on your own hardware, with nothing leaving your network.
Get started
- Learn more at withcoral.com
- Install guides for desktop app and to self-host for your team