Covenant gives AI systems scoped authority, durable state, runtime control, and verifiable evidence for work that spans tools, sessions, and organizations. Teams can define what a system may do, restore its state after interruption, and trace consequential actions through durable records.

Website · Documentation · Core architecture · Live sandbox

Covenant Timeline gives models and applications a portable representation of events, intervals, plans, forecasts, and corrections. It distinguishes when a fact was valid from when it was observed or recorded, so late evidence can update current reasoning without changing earlier answers.

Typed queries return bounds, temporal relations, or contradictions with proof receipts that another process can verify. The same temporal state can be shared across models, runtimes, languages, and workflows.

Timeline v0alpha3 is available in @covenant-org/timeline@0.0.0-alpha.2:

npm install --save-exact @covenant-org/timeline@0.0.0-alpha.2

The API is defined by Draft RFC 0009. Review the alpha.2 release and Timeline release status for supported APIs, distribution channels, and maturity.

Explore Timeline · Read the vision · Integrate a model

flowchart LR
    intent["Human intent and operator policy"] --> core["Covenant control plane"]
    app["AI application or runtime"] <--> core
    core --> authority["Capabilities"]
    core --> continuity["Durable state"]
    core <--> timeline["Temporal state and proofs"]
    core --> evidence["Audit, provenance, attestations and receipts"]
    tools["Tools and external services"] <--> core

Covenant manages authority, continuity, and controlled execution. Timeline provides queryable temporal state and conclusions with machine-checkable proofs. Trust Layer produces deterministic policy decisions and receipts. The plugin, skill, and SDK surfaces connect these capabilities to applications, agent runtimes, and operator workflows.

• Software delivery: constrain agent authority, retain state through restarts, and preserve the evidence behind review and release decisions.
• Long-running operations: coordinate work that spans workers, tools, and approval boundaries with a durable audit trail.
• Research and analysis: represent observations, validity periods, competing hypotheses, and later corrections as inspectable temporal state.
• Governed workflows: combine domain-owned policy and evidence with explicit capabilities, provenance, and deterministic verification.
• AI platforms: provide shared control and evidence services across models, runtimes, and application teams.

Provider-neutral interfaces support optional adapters for identity, inference, data, payments, settlement, and evidence systems. The capability index and integration documentation describe each implemented surface.

1. Explore the live sandbox.
2. Run the Covenant quick start against a bounded local workflow.
3. Choose one bounded workflow and define the authority, state, and evidence it must preserve.
4. Integrate through the relevant SDK, plugin, skill, or model interface.
5. Test revocation, restart recovery, and evidence replay under the deployment policy.

Model and runtime teams can start with the Timeline model interface and executable temporal pilot. Independent implementers can use RFC 0009 and the conformance policy.

Covenant is alpha infrastructure for evaluation, integration, and contribution. The local control plane, capability lifecycle, durable memory, audit and provenance surfaces, SDKs, and settlement adapters are implemented and tested. Timeline v0alpha3 ships in @covenant-org/timeline@0.0.0-alpha.2 for technical evaluation; its API remains Draft.

Before production or high-stakes use, review the selected component's release status, security model, and deployment requirements.

Open focused issues and proposals in the relevant repository. Security researchers should use that repository's private vulnerability-reporting channel.