bighub 0.1.0b2

Better Decision SDK for IT agent actions with BIGHUB.

BIGHUB SDK

Better decisions for IT agent actions.

Beta release: this is the first public Better Decision SDK beta.

BIGHUB helps AI agents and CI/CD workflows make better IT decisions before execution.

It takes a proposed IT action, builds a Decision Packet, runs DecisionBrain, and returns execution guidance:

• proceed when appropriate
• pause for review
• ask for more context
• advise not to run
• suggest a real better_action only when the backend actually produced one

pip install --pre bighub

For future stable releases, use:

pip install bighub

Python 3.9+. Single dependency: httpx.

---

Quick Start

from bighub import Bighub


def run(action_to_run):
    raise NotImplementedError("Connect this to your executor.")


bighub = Bighub(api_key="your_api_key")

decision = bighub.decide(
    action="Roll out billing-api v3.12 to production",
    context={
        "system": "kubernetes",
        "environment": "production",
        "service": "billing-api",
        "ticket": "CHG-4401",
    },
)

if decision.needs_review:
    decision.request_review()
elif decision.needs_more_context:
    print("More context required:", decision.reason)
elif decision.should_not_run:
    print("Do not run:", decision.reason)
elif decision.can_run:
    action_to_run = decision.better_action or decision.proposed_action
    run(action_to_run)

bighub.close()

Recommended flow:

proposed IT action -> Decision Packet -> DecisionBrain -> better_action when real -> execution mode and flags -> review or context when needed

---

When to use BIGHUB

Use BIGHUB before risky actions such as:

• GitHub Actions, GitLab CI, or Jenkins deploys
• terraform apply
• Argo CD sync or Kubernetes / OpenShift rollout
• Okta or IAM access changes
• credential rotation
• incident updates in Slack
• security gates after Trivy / Syft / SBOM signals
• production actions that may need review or additional context

If your workflow only reads data or performs harmless lookups, you probably do not need BIGHUB.

---

What a Decision returns

High-value fields most workflows care about:

• better_action: a real backend-produced alternative when available, otherwise None
• mode: execution guidance such as autonomous, constrained, review, blocked, or needs_context
• can_run, needs_review, needs_more_context, should_not_run: direct operational flags
• risk: top-level risk score when provided
• decision_path: routing or evaluation path when provided
• packet: the DecisionPacket used for the decision
• brain: the DecisionBrainResult with reasoning summary, confidence, expected regret when provided, and related signals
• selected_model: present only when the backend actually selected one

For a smaller stable surface, use decision.brief():

brief = decision.brief()

if brief.can_run:
    run(brief.recommended_action)
elif brief.needs_review:
    print("Review required:", brief.reason)

---

Honest behavior

The SDK is intentionally strict about what it does and does not invent:

• better_action is None unless BIGHUB returned a genuinely distinct recommendation
• selected_model is None unless the backend actually selected one
• decision_path and model-selection fields stay empty when the backend did not provide them
• outcomes are optional and not required for a first integration
• legacy BighubClient and client.actions.evaluate(...) remain available for existing integrations

If a backend decision_packet has no packet_sha256, the SDK computes a stable local hash and marks it with packet.packet_sha256_is_local=True.

---

Optional: outcome reporting

If you want a learning loop later, report what happened after execution:

decision.report_outcome(
    status="completed",
    evidence={"deployment_id": "dep_123"},
)

Outcome reporting is optional. The primary integration wedge is still the decision before execution.

---

CI/CD and system integrations

You can connect system evidence so BIGHUB sees recent operational context before deciding.

For lower-level resources, BighubClient remains available:

from bighub import BighubClient

client = BighubClient(api_key="your_api_key")

client.systems.save_connection(
    "gitlab",
    {
        "base_url": "https://gitlab.com",
        "gitlab_token": "glpat_...",
        "project_id": "123",
    },
    display_name="GitLab production",
)

client.systems.update_poll_schedule("gitlab", enabled=True, interval_seconds=300)
client.systems.poll("gitlab")
world = client.systems.world_state()

client.close()

System providers currently supported by the SDK/API include okta, slack, github, datadog, sentry, aws_cloudtrail, terraform, kubernetes, argocd, gitlab, jenkins, azure, prometheus, grafana, and openshift.

Poll snapshots and histories are redacted before persistence and exposure. They provide best-effort operational context; they are not a replacement for production monitoring.

Typical uses:

• gate a GitHub Actions or GitLab deploy with recent operational context
• inspect Kubernetes, Argo CD, Prometheus, or Grafana state before rollout
• apply IAM or Okta changes with better review posture
• incorporate security and incident signals before execution

---

Compatibility

The modern entrypoint is bighub.decide(...).

For existing integrations, legacy BighubClient, client.actions.submit(...), and client.actions.evaluate(...) remain supported.

---

Links

• bighub.io
• GitHub repository
• Python examples
• OpenAI adapter
• MCP server
• PyPI - bighub
• PyPI - bighub-openai

---

License

MIT