Roadmap

Built in the open.

Everything we're shipping — visible, versioned, and committed to. Shipped work carries a green pill. Active build is amber. Future work is gray. No vaporware; if it's here, it has a real owner and a real milestone.

Shipped live today Building active development Planned scoped · not started
Now
Shipped · 2026 Q1 – Q2
Shipped
Dream Engine · 9 strategies
Synthesize, pattern extract, contradiction detect, compress, associate, validate, evolve, forecast, reflect.
Shipped
LongMemEval 94.6%
Third-party benchmark. 473/500 under byte-exact upstream GPT-4o judge. Reproducible methodology published.
Shipped
Multi-agent namespaces
Scoped memory for every agent in a crew. No context bleed.
Shipped
RBAC
Role-based access on every memory, namespace, and channel.
Shipped
Webhooks
HMAC-signed outbound events on memory + dream lifecycle.
Shipped
WebSocket events
Real-time memory streaming for live agents and UI.
Shipped
Agent Card (A2A)
Native agent-to-agent discovery over the A2A protocol.
Shipped
Graph queries
Traverse entities, relations, and edges across your knowledge graph.
Shipped
Self-host docker
Single image. SQLite-backed. Air-gapped friendly.
Shipped
80+ integrations
Slack, Linear, Stripe, Shopify, Zapier, Make, and the long tail.
Shipped
Obsidian sync
Two-way vault import with wikilinks, tags, and folder structure preserved.
Shipped
ChatGPT import
One-click import of the ChatGPT export bundle. Threads become memories.
Next
Building · 2026 Q3
Building
Native MCP server binary
Single-binary MCP server. Drop-in for Claude Desktop, Zed, and every MCP-native client.
Building
Python/JS SDKs (beta → GA)
Typed, idiomatic. Auto-generated from the OpenAPI spec. Promoted to GA this quarter.
Building
CrewAI / LangChain / LlamaIndex adapters
First-class integration with the three most-used agent frameworks.
Building
Memory lineage UI
See exactly how a memory was derived — source, dreams applied, confidence trail.
Building
Temporal reasoning
Belief-change detection. Surface when and why your knowledge updated.
Building
Memory confidence scores
Calibrated probability on every recall. Ranked queries by reliability.
Building
Bulk import/export tools
CLI and REST for multi-GB transfers. Background jobs with progress.
Building
OpenAPI spec public
Versioned spec at the well-known URL. Auto-generated client stubs.
Building
Postman collection
Every endpoint, every example, keyed to the free tier. One-click fork.
Building
Sandbox environment
Zero-setup playground with synthetic data. No credit card. No install.
Later
Planned · 2026 Q4 +
Planned
SOC 2 Type II
Audit in flight. Enterprise security posture with independent attestation.
Planned
HIPAA compliance
BAA-ready. Healthcare and clinical agents with PHI-safe namespaces.
Planned
Multi-modal memory
Images, audio, and video as first-class memory primitives. Same graph, same dreams.
Planned
Proprietary embedding model
Trained end-to-end for consolidation tasks. Better recall at a fraction of dim.
Planned
Consolidation foundation model
Fine-tuned model that runs the Dream Engine offline. Open-weights release planned.
Planned
EU data residency
Regional clusters. Full GDPR sovereignty. Frankfurt and Dublin first.
Planned
Memory benchmark dataset
A proprietary, open-sourced eval for memory systems. Because LongMemEval isn't enough.
Planned
Memory marketplace
Portable, permissioned knowledge bundles. Curated graphs, shared across teams.
Commitment

This roadmap is a promise, not marketing. If a commitment slips, we say so here first. If a feature ships, the pill flips to green the same day. Changes are tracked in the changelog. Open a feature request on GitHub or drop into Discord to influence what moves first.