Everruns

Everruns is an MIT-licensed Rust framework for putting AI agents directly inside an application, plus an optional shared platform for teams that need durable sessions and operational controls. The project keeps the application-facing loop explicit: define an agent, attach a model and typed tools, create a session, and run a turn. Its platform layer is separate, so a Rust service can begin embedded and later add a self-hosted or hosted control plane when it needs remote execution, channels, and persistence.

What it provides

• Rust-native agent loop: the everruns crate exposes agents, sessions, provider configuration, typed tools, events, cancellation, and lifecycle hooks instead of hiding application behavior behind a prompt-only workflow.
• Deliberate capability boundaries: applications opt into filesystem, shell, web, Lua, and MCP capabilities. That is useful when a tool needs a narrower permission set than the model itself.
• Durable platform option: the self-hosted platform and Everruns Cloud add a web UI, stateless workers, a remote API, durable execution, organization controls, budgeting, and evaluation.
• Multiple delivery channels: the official documentation lists web chat, Slack, A2A, webhooks, schedules, voice, HTTP, and MCP as platform publishing targets.
• Current release direction: version 0.48.0, published on 2026-10-13, moved platform sessions to durable actors and introduced a stable Agent API and consolidated web-search tooling. The release also retires older api_endpoint channel routes, so existing clients should read its upgrade notes.

A practical workflow

Start in a Rust service with the framework quickstart: configure a model provider, implement one typed function tool, and run a small task in a new session. Keep the tool body responsible for authorization and validation rather than granting a broad shell or network capability. After the agent behavior is predictable, choose the platform only if the job needs crash recovery, a shared worker pool, channel delivery, or organization-level budgets. For example, a support triage agent can begin in-process, then move to the platform when Slack messages, scheduled follow-ups, and an auditable session history become production requirements.

Pricing and deployment

The framework is open-source under the MIT license and can be embedded or self-hosted. Everruns also offers a cloud product, but the public project pages do not state a fixed hosted price table. Budget for the model provider, storage, workers, search services, and any chosen hosting or cloud plan rather than assuming the open-source license covers those operating costs.

How it compares

Mastra is a TypeScript-oriented agent framework, while Everruns targets Rust applications and makes typed Rust tools its starting point. LangGraph is useful when graph-based orchestration is central. Everruns is most distinctive for teams that want an embedded Rust API first and a later path to durable, multi-channel operations without changing the agent's core session model.

Risks and next step

Everruns is under active development, and its October release includes a breaking migration from older channel routes. Durable execution does not remove the need for idempotent tool design, scoped credentials, rate limits, and human approval around consequential actions. Read the framework quickstart, run one read-only typed tool locally, then review the 0.48.0 changelog before adopting the platform API.

FAQ

Can I use Everruns without operating its platform?

Yes. The framework is designed to be embedded in a Rust application. The platform is an optional path for shared runtime and production operations.

Does the platform make agent tools safe by default?

No. It supplies runtime controls and documented capability boundaries, but application code still needs least-privilege credentials, validation, and approval rules.

When should I use the Agent API?

Use it when another application needs to create sessions, stream replies, answer agent questions, or approve tools through a stable service interface. Check the current release notes because the older channel route format was retired.