Features
Everything Novyx ships — memory, rollback, audit, and the tools built on top.
Memory Store & Recall
Store observations with tags, importance scores, and context. Recall semantically — ask a question, get ranked results in milliseconds.
- ✓Sub-100ms semantic recall
- ✓Importance scoring and tag filtering
- ✓Survives restarts, crashes, redeployments
- ✓Share across agents with Context Spaces
User prefers async communication
Project deadline is March 15
Dark mode enabled across all apps
Weekly standup moved to Tuesdays
User prefers async communication
Weekly standup moved to Tuesdays
Dark mode enabled across all apps
Magic Rollback
Point-in-time restore for your agent's memory. Preview what will change before executing. Like git reset for agent state.
- ✓Restore to any point in time
- ✓Preview changes before executing
- ✓Compensation plans for side effects
- ✓AI-suggested rollback points
Knowledge Graph
All plans
Your agent doesn't just remember facts — it understands relationships. Store structured triples, traverse entities, and build smarter agents.
- ✓Subject–predicate–object triples
- ✓Entity traversal and auto-dedup
- ✓Cascade deletes for clean removal
- ✓Force-directed graph visualization
Cryptographic Audit Trail
Every memory operation is SHA-256 hashed and timestamped. Tamper-proof, exportable, and independently verifiable.
- ✓SHA-256 cryptographic hashing
- ✓Hash-chained operations
- ✓CSV/JSON export for compliance
- ✓RSA-signed execution traces
| Timestamp | Operation | Agent | Hash | Status |
|---|---|---|---|---|
| 10:18:42 | CREATE | agent_alpha | a3f8c2...e91b | |
| 10:18:45 | UPDATE | agent_alpha | 7d2e1f...b4a3 | |
| 10:19:01 | CREATE | agent_beta | e5c9a7...2f18 | |
| 10:22:30 | ROLLBACK | agent_alpha | 91b3d4...c7e2 | |
| 10:22:31 | DELETE | system | f4a2b8...3d91 |
Execution Traces
All plans
Record every step of your agent's reasoning chain. Thought, action, observation, output — cryptographically signed and hash-chained.
- ✓THOUGHT → ACTION → OBSERVATION → OUTPUT
- ✓RSA-4096 signed, hash-chained
- ✓Independently verifiable
- ✓Integrates with Replay time-travel
User asked about Q4 priorities — checking memory...
nx.recall("Q4 priorities", limit=5)
3 memories found: budget review, team expansion, launch date
Based on your notes, the Q4 priorities are...
Compensation Webhooks
NewAll plans
When a rollback detects side effects in your traces — API calls, webhook fires, external writes — Novyx generates a compensation plan and notifies you.
- ✓Automatic side-effect detection
- ✓Reversal plan in execution order
- ✓Slack and Discord native formatting
- ✓Acknowledge actions as completed/failed/skipped
triggers scan
find side effects
reversal actions
Slack / Discord
Rollback compensation: 3 actions need attention
Eval — Memory Health Scoring
New in v3.0.0Score your agent's memory health, detect regressions before they reach users, and gate CI/CD deployments on memory quality. The only SDK that tells you if your agent is getting dumber.
- ✓Composite health score (0-100) from recall, drift, conflicts, staleness
- ✓CI/CD quality gate — block deploys below your threshold
- ✓Baseline regression testing — save queries, detect when results degrade
- ✓Drift analysis — track topic churn, importance shifts over time
- ✓Eval history with per-tier retention up to 365 days
Health Score
500 memories · 0 conflicts · 140 stale
$ curl -X POST /v1/eval/gate -d '{"min_score": 80}' --fail-with-body
exit 0 — deploy proceeds
Runtime v2 — Agent Lifecycle
New in v2.4.0All plans
First-class primitives for orchestrating autonomous agents. Create agents that persist across sessions, assign goal-oriented missions, save checkpoints, and intervene when it matters.
- ✓Agents — create, inspect, delete. Persistent identity across sessions
- ✓Missions — goal-oriented task orchestration with pause, resume, cancel
- ✓Capability Packs — define which tools an agent can use. Policy at the orchestration layer
- ✓Checkpoints — save and restore agent state. Rollback to any checkpoint
- ✓Supervisor Interventions — inspect and override agent decisions in real-time
deploy-agent wants to run: kubectl apply -f prod/rollout.yaml
More Core capabilities
Cortex
NewAutonomous memory intelligence. Auto-merge duplicates, boost important memories, decay noise, and surface patterns.
Explore Cortex →Webhooks
All plans
Real-time event notifications. HMAC-signed payloads with Slack and Discord native formatting.
View docs →Teams
All plans
Invite your team to a shared memory namespace. Owner, admin, member, and read-only roles.
View docs →Context Spaces
All plans · sharing on Pro+
Agent A learns it, Agent B knows it. Shared memory across your entire agent fleet.
View docs →Milestones
All plans
Tag meaningful points in your timeline — deployments, model swaps, prompt changes. Diff before vs. after.
View docs →Hygiene
NewContext hygiene for agentic coding. Auto-persists Claude Code sessions across /compact, /clear, and restarts via hooks. Health scoring included.
View docs →Agent SDK
NewBuild agents with built-in memory. Auto-recall before every response, auto-remember after. Works with OpenAI, Anthropic, or any LiteLLM model.
View docs →Control
NewAction governance for AI agents. Policy engine, approval flows, and cryptographic audit trail for every tool call that touches the real world.
View docs →Sentinel Circuit Breaker
Automatic protection against cascading failures. Configurable fail-open/fail-closed modes. Free on every plan.
View docs →Replay — Time-Travel Debugging
All plansSee exactly what your agent knew when it made a decision. Timeline, snapshots, diffs, counterfactual recall, and drift analysis.
Production-ready from day one
p50: <100ms
Median recall latency
p95: <200ms
95th percentile recall
Up to 99.9%
uptime SLA
Your data is yours — export anytime, no lock-in.
Safety & reliability built in
Execution Traces
RSA-4096 signed record of every agent step. Independently verifiable proof of what your agent did and why.
Circuit Breaker
Automatic protection against runaway loops, data exfiltration, and suspicious operations. Free on every plan.
Anomaly Detection
Behavioral baselines detect unusual agent activity. Get alerted before users notice.
Conflict Resolution
Vector clocks handle concurrent writes. Last-write-wins, reject, or flag for manual review.
What developers are building
Cross-Session Recall
A scheduling agent that auto-recalled a contact's timezone and meeting preferences in a brand new session — without being told.
30-Second Recovery
An agent that stored bad data for 3 days. One rollback command. 30 seconds. Clean state.
Ready to try it?
Free tier. No credit card. Start in 30 seconds.