Python SDK

pip install novyx — 85+ methods for persistent memory, rollback, audit, knowledge graph, eval, governance, and more.

Breaking change in 3.3.0

nx.create_agent() now requires provider and model keyword arguments. The previous OpenAI default was removed in Phase 3 of the governance shipment. See the novyx-agent 2.0 upgrade guide for the same change in the higher-level Agent class.

New in 3.4.0 — nx.submit_action()

Typed wrapper around POST /v1/actions for the main cloud governance flow. Distinct from the legacy nx.action_submit() which targets a separate Control instance via control_url. See the Control section for the recommended pattern.

Installation

pip install novyx          # Sync client
pip install novyx[async]   # Async client (httpx)

Quick Start

from novyx import Novyx

nx = Novyx(api_key="nram_...")

# Store a memory
result = nx.remember("User prefers dark mode", tags=["preferences"])

# Recall semantically
memories = nx.recall("user preferences")

# Rollback to a point in time
nx.rollback("2 hours ago")

Async Client

from novyx import AsyncNovyx

nx = AsyncNovyx(api_key="nram_...")
result = await nx.remember("User prefers dark mode")
memories = await nx.recall("user preferences")

Core Methods

Memory

Method	Description
`nx.remember(observation, *, tags, importance, context, agent_id, space_id, ttl_seconds)`	Store a memory
`nx.recall(query, *, tags, limit, min_score)`	Semantic search across memories
`nx.memories(*, tags, limit, offset)`	List memories with optional filtering
`nx.memory(memory_id)`	Get a single memory by ID
`nx.forget(memory_id)`	Delete a memory
`nx.supersede(old_id, new_id)`	Mark one memory as superseding another
`nx.stats()`	Memory count, tag distribution, importance stats
`nx.memory_health()`	Composite health score (0–100)

Rollback & Audit

Method	Description
`nx.rollback(target)`	Undo operations after a timestamp (supports "2 hours ago")
`nx.rollback_preview(target)`	Preview what a rollback would change
`nx.rollback_history(limit=50)`	List past rollback operations
`nx.audit(*, limit, offset)`	View the cryptographic audit trail
`nx.audit_verify()`	Verify hash chain integrity
`nx.audit_export(format="csv")`	Export audit log as CSV or JSON

Knowledge Graph

Method	Description
`nx.triple(subject, predicate, object)`	Add a subject-predicate-object triple
`nx.triples(*, subject, predicate, object)`	Query triples
`nx.delete_triple(triple_id)`	Delete a triple
`nx.entities(*, limit, offset)`	List entities in the graph
`nx.entity(entity_id)`	Get entity details
`nx.delete_entity(entity_id)`	Delete an entity
`nx.graph(*, limit)`	Get graph edges

Context Spaces

Method	Description
`nx.create_space(name, description)`	Create an isolated memory namespace
`nx.list_spaces()`	List all spaces
`nx.get_space(space_id)`	Get space details
`nx.update_space(space_id, *, name, description)`	Update a space
`nx.delete_space(space_id)`	Delete a space
`nx.space_memories(space_id, *, limit, offset)`	List memories in a space
`nx.share_context(*, tags, target_tenant)`	Share tagged memories with another tenant
`nx.shared_contexts()`	List shared contexts

Replay (Pro+)

Method	Description
`nx.replay_timeline(*, limit, offset)`	Full timeline of memory changes
`nx.replay_snapshot(at, *, limit)`	Point-in-time memory snapshot
`nx.replay_memory(memory_id)`	Full lifecycle of a single memory
`nx.replay_recall(query, at, *, limit)`	"What would recall have returned at time X?"
`nx.replay_diff(from_ts, to_ts)`	Compare memory state between two timestamps
`nx.replay_drift(from_ts, to_ts)`	Analyze how memory composition drifted

Cortex (Pro+)

Method	Description
`nx.cortex_status()`	Check Cortex analysis status
`nx.cortex_config()`	Get current Cortex configuration
`nx.cortex_update_config(**kwargs)`	Update Cortex settings
`nx.cortex_run()`	Trigger autonomous memory maintenance
`nx.cortex_insights(*, limit, offset)`	Get AI-generated insights

Eval

Method	Description
`nx.eval_run(*, min_score)`	Run memory health evaluation
`nx.eval_gate(min_score)`	CI/CD quality gate — fails if health below threshold
`nx.eval_history(*, limit, offset)`	Past evaluation results
`nx.eval_drift(*, days)`	Memory drift analysis over time
`nx.eval_baseline_create(query, expected)`	Create a recall baseline
`nx.eval_baselines()`	List baselines
`nx.eval_baseline_delete(baseline_id)`	Delete a baseline

Control — Actions & Approvals

Method	Description
`nx.submit_action(action, params=None, *, agent_id=None)`	Recommended. Submit an action to the main cloud governance flow (`POST /v1/actions`). Evaluates against built-in + custom YAML policies. Returns status `allowed`, `blocked`, or `pending_review`. New in 3.4.0.
`nx.action_status(action_id)`	Check action status, including post-approval result.
`nx.action_list(status=None, *, limit=None)`	List recent Control actions.
`nx.policy_check(agent_id=None, connector=None, operation=None)`	Read the active policy profile.
`nx.list_approvals(*, limit=50, status_filter=None)`	List pending action approvals.
`nx.approve_action(approval_id, *, decision="approve", reason=None, approver_id=None)`	Approve or deny a pending action.
`nx.explain_action(action_id)`	Get the full causal chain for an action — policies, approval, memories, audit.
`nx.action_submit(connector, operation, payload)`	Legacy. Submit a `strata.action.v0` envelope to a separate Control instance. Requires `control_url` set on the client. Use `submit_action()` for the main API.

Control — Custom Policies (new in 3.3.0)

Method	Description
`nx.create_policy(name, *, rules, description="", step_types=None, whitelisted_domains=None, enabled=True, agent_id=None)`	Create or update a custom YAML/dict policy. Upserts on existing name.
`nx.list_policies(*, agent_id=None)`	List active policies. Pass `agent_id` to also include agent-scoped overrides.
`nx.get_policy(policy_name, *, agent_id=None)`	Fetch one policy's full configuration. Scope-aware.
`nx.update_policy(policy_name, *, rules, description="", ...)`	Replace an existing policy's rules. Increments `version`.
`nx.delete_policy(policy_name, *, agent_id=None)`	Soft-delete (disable) a custom policy. Built-ins cannot be deleted.

All five accept optional agent_id for agent-scoped policies (Pro+).

Control — Governance Dashboard (new in 3.3.0)

Method	Description
`nx.governance_dashboard(*, window="7d", bucket=None)`	Aggregated stats: totals, violations by policy, by agent, time-series. Window: `24h`, `7d`, `30d`. Starter+.
`nx.agent_violations(agent_id, *, limit=50, since=None, until=None)`	Per-agent violation history from the audit chain. Starter+.

Memory Drafts

Method	Description
`nx.draft_memory(observation, *, tags, importance)`	Create a draft (not yet committed)
`nx.memory_drafts(*, status, branch_id)`	List drafts
`nx.draft_diff(draft_id)`	Compare draft to current state
`nx.merge_draft(draft_id)`	Commit a draft to memory
`nx.reject_draft(draft_id, *, reason)`	Reject a draft

Streaming

Method	Description
`nx.stream(event_types)`	Subscribe to real-time SSE memory events
`nx.stream_status()`	Check stream connection status

Error Handling

from novyx import Novyx, NovyxError

nx = Novyx(api_key="nram_...")

try:
    nx.remember("something")
except NovyxError as e:
    print(f"Error: {e.status_code} — {e.message}")

Environment Variables

Variable	Description
`NOVYX_API_KEY`	Your API key (alternative to passing in constructor)
`NOVYX_BASE_URL`	Override the API base URL (default: `https://novyx-ram-api.fly.dev`)

Installation​

Quick Start​

Async Client​

Core Methods​

Memory​

Rollback & Audit​

Knowledge Graph​

Context Spaces​

Replay (Pro+)​

Cortex (Pro+)​

Eval​

Control — Actions & Approvals​

Control — Custom Policies (new in 3.3.0)​

Control — Governance Dashboard (new in 3.3.0)​

Memory Drafts​

Streaming​

Error Handling​

Environment Variables​