Reproducible bundles for autonomous actors.
Immutable, content-addressed, Sigil-signed bundles. Everything an autonomous actor needs to execute, packaged as a single attestable artifact.
What it is
A Seal Bundle contains everything required to instantiate a specific class of actor at a specific version: behavioral specifications, tool references, model references, memory schemas, and — for embodied actors — firmware, model weights, sensor calibrations, behavior trees. Every Bundle is content-addressed and signed by the issuer's Sigil. Two actors with the same Bundle hash are identical in every behaviorally relevant respect. The unit of reproducible autonomy.
The problem
Production agents today are rarely reproducible. The prompt was edited in a UI. The tool list was changed via a config push. The model version was upgraded silently. Six months later, a regression appears, and there is no way to recover the exact actor that ran in production a quarter ago. For embodied actors, the problem is sharper: firmware, model weights, and behavior tree changes flow through different channels with no unified versioning.
Capabilities
A bundle definition
bundle "acme/triage" {
version = "3.2.1"
behavior {
type = "agent"
prompt = file("./prompts/triage.md")
tools = [
"mcp://zendesk@^2.0.0",
"mcp://kb-search@1.4.x",
]
}
model {
primary = "anthropic/claude-sonnet-4-6"
fallback = "openai/gpt-5"
ceiling = { tokens = 50000, cost_usd = 0.50 }
}
memory {
schema = file("./schemas/triage-memory.yaml")
retention = "30d"
}
capabilities {
requires = [
"zendesk:tickets:assigned:read",
"zendesk:comments:assigned:write",
"kb:articles:public:read",
]
}
}Open and commercial
Seal is delivered through two surfaces: an open-source reference in Lattice Runtime (MPL 2.0), and a managed, enterprise-grade delivery in Visca Cloud. Features in the open never move behind the paywall.
Open
MPL 2.0 · self-hostable · foundation-track governance
Commercial
Hosted SaaS · self-managed · hybrid · air-gapped
ALM · same pillar
Data model as code.
The declarative primitive for your data model. Agents write it as source; humans edit it as a spreadsheet. Plan, apply, drift-detect.
Autonomy-aware runtime.
Runs autonomous workloads across cloud, edge, on-device, and on-robot — with first-class awareness of tokens, models, memory, and embodiment.
The developer surface.
Where developers build autonomous systems. CLI, SDKs, local runtime, eval harness, replay, documentation, templates — one surface.
Across the platform
Cryptographic identity for autonomous actors.
Issues, verifies, and revokes the cryptographic identity every autonomous actor holds — bound to principal, runtime, and lineage from the moment of instantiation.
Zero-trust access for autonomous actors.
Brokers every access from an autonomous actor to a consequential resource — identity-bound, scoped, time-limited, audited. Long-lived credentials disappear.
The autonomic fabric.
Discovery, mTLS, and routing across every autonomous actor and every resource an actor reaches — rooted in Sigil identities, not hardcoded URLs.
The system of record.
A queryable knowledge graph of every autonomous actor in the estate — and the append-only, cryptographically chained audit of every action taken.
The autonomy economy is shipping
Build, run, and trust autonomous systems on an integrated platform — software, embodied, and hybrid. Open foundation. Enterprise cloud. No license rug-pulls, ever.