Deployments as runtime contracts

Ship agents like you ship everything else.

Turn agent code into a versioned, reproducible deployment. Runtime metadata, tools, skills, knowledge, policies, and environment requirements — all declared in one place.

Agent code
↓
Build
↓
Versioned image
↓
Deployed

Agent code is not a production boundary.

Production agents need more than source code and prompts. They need a deployable contract — how they run, which tools and skills they can reach, which knowledge they can access, where they execute, and which policies apply.

Agent code
↓
Deployment contract
↓
Versioned runtime
↓
Controlled execution

Everything the runtime needs to execute an agent safely.

A sypho deployment describes the operational shape of an agent. It connects code, runtime, tools, skills, knowledge, policies, and target environments into one versioned unit.

Nothing implicit
Agent version
Runtime and entrypoint
Tool definitions
Skills
Knowledge scope
Environment variables
Secrets references
Runner target
Runtime policies
Promotion state
In practice

One file says what the agent is allowed to be.

deployment manifest
agent: claims-assistant
version: 1.2.0
entrypoint: claims:agent
tools:
  - lookup_policy
  - create_case
skills:
  - claims-playbook
  - escalation-guide
structured_data:
  policy-docs: read
  claims: read_write
deployment:
  type: production
  budget_usd: 5.00

From local build to a running production version.

Build
→
Deploy
→
Run
→
Observe

Build a versioned image. Deploy it to the environment it belongs in. Run it. Observe every execution.

Build

Produce a versioned image and manifest from your agent code — tools, skills, and structured-data dependencies included.

Deploy

Register the manifest and target a dev, staging, or production environment.

Observe

Every run streams its events, tool calls, and results back to the control plane as it happens.

Roll out agent changes deliberately.

Agent behavior can change when prompts, tools, structured data, or models change. sypho lets you target a new version at dev or staging before it reaches production traffic.

Build new version
↓
Deploy to staging
↓
Run test cases
↓
Review traces
↓
Deploy to production
compare versions before rollout
validate tool access
test structured-data scope
inspect traces
reduce release risk

Deployment that holds up under real traffic.

Reproducible releases

Package runtime configuration with the agent version, not scattered across scripts and environments.

Environment-aware execution

Target local, private cloud, customer cloud, on-prem, or sypho-managed runners.

Controlled access

Attach tools, skills, structured data, secrets, and budgets intentionally per deployment.

Safer rollout

Validate a new version in staging before it reaches production.

Full traceability

Connect every production run back to the deployment version and runtime configuration that produced it.

Deploy agents with production discipline.

Request access and help shape the runtime layer for agentic software.

Request access