Roadmap === This is the public roadmap for Signet, updated manually as priorities shift. It captures what we're actively building, what's committed next, and what we're still shaping. For implementation sequencing, dependency contracts, and status truth, see [Spec Index](./docs/specs/INDEX.md). Status markers: `[done]` shipped, `[wip]` in progress, `[next]` planned, `[stub]` intentionally tracked but not fully planned, `[idea]` exploring. Current Focus (0.124.x) --- - [wip] Public positioning and documentation correctness - align README, docs, and website language around portable agent state, local ownership, source truth, and behavioral context portability - [wip] Source-native substrate - preserve source artifacts as ground truth, derive semantic memory with provenance, and make claims rebuildable from their evidence - [wip] Source events as agent triggers - treat sources as live event surfaces, not only passive recall inputs: events should carry source truth, policy, context, and a result artifact - [wip] Skills over substrate - keep deterministic substrate in Signet and move repeated cognition into loadable, inspectable, portable skills with their own working trails - [wip] Proving-ground loop - use one source ecosystem, one recurring job-to-be-done, one visible metric, and one tight loop where bad memory visibly hurts - [wip] Reliability and performance as product integrity - reduce daemon fragility, runtime cost, and resource waste because local ownership should not push access debt back onto users - [wip] Authority, permissions, and secrets boundary - agent-blind secrets, RBAC, token policy, rate limits, capability scoping, and source-backed authority across protocols and MCP tools Planned Next --- - [next] Universal source layer - make it easy for an agent to connect Obsidian, repos, transcripts, Discord, Fitbit, servers, devices, and other sources through one common source-artifact contract - [next] Pipeline V3 as skills over substrate - run scheduled and event-triggered skill sessions over source artifacts, transcripts, failures, and prior decisions; emit proposed graph, memory, identity, skill, source, or authority changes with provenance - [next] Dreaming as background semantic QA - judge memory maintenance by downstream behavior: fewer repeated investigations, fewer repeated mistakes, cleaner source-backed memory, better entity density, and visible mutation diffs - [next] Obsidian proving ground - harden the clipping-to-literature workflow as the first bathtub: raw source, adversarial intake, provenance, synthesis, backlinks, indexes, operations logs, skill formation, and user taste in one inspectable loop - [next] Hermes Agent production hardening - validate real install flows, gateway session routing, and cross-platform memory continuity; borrow the stronger skill layer where it fits Signet - [next] Capability-scoped tool permissions - distinguish read-only, destructive, credentialed, and approval-gated actions across harnesses and MCP tools - [next] Signet Cloud baseline inference and sync - keep the open-source product complete while exploring managed sync, background inference, and stronger embedding models as hosted help - [next] Rust daemon parity + primary runtime cutover - parity first, cutover second, with divergence logging, rollback paths, and performance wins validated before the TypeScript daemon gives way Strategic Stub Backlog (tracked, planning incomplete) --- - [stub] Behavioral context portability / OAS working draft - define export/import expectations for identity, memory, provenance, skills, source metadata, and secrets metadata without tying the standard to one harness - [stub] Distributed harness and multi-remote orchestration - [stub] Remember/Recall skill parity refresh with current schema - [stub] Per-user neural query surface - train a disposable small model on a user's artifact history as a future query layer, while keeping source artifacts underneath as ground truth - [stub] Deep memory search as an optional escalation path - multi-agent or LLM-assisted search only when primary retrieval cannot answer with enough confidence - [stub] MCP CLI bridge + usage analytics in dashboard - reference direction: MC Porter-style command visibility - [stub] Git-based marketplace monorepo for skills + MCP servers - GitHub-authenticated PR publishing/reviews, JSON review artifacts - [stub] Adaptive skill lifecycle - usage tracking, maintenance suggestions, and procedural memory links - [stub] Cryptographic identity roadmap - agent passports, artifact signing, and verifiable identity without requiring a specific chain or vendor identity provider - [stub] Plugin API + app ecosystem integrations - dashboard and CLI surfaces for third-party integrations - [stub] Unified realtime constellation / embedding / entity viewer - replace slow current 3D path and make provenance inspectable by default - [stub] Dashboard IA refactor - settings as standalone page, breadcrumb-first navigation model - [stub] Post-install behavior migration audit - ensure daemon/CLI own critical install flows consistently Recently Shipped --- - [done] Readable changelog generation and release-note correctness - release entries now use the bumped version and include a highlights layer plus exact release ledger - [done] Workspace source checkout sync - managed source checkout, clean-branch update behavior, and watcher ignore coverage - [done] Connector expansion and hardening - Hermes Agent integration, OpenClaw runtime hardening, and request normalization - [done] Recall surface alignment - CLI, MCP, SDK, and hook recall paths now share more of the same filter and response contract - [done] Documentation site quality pass - section-aware docs search, normal markdown link rewriting, TOC fixes, and clearer contributor/user workflow docs - [done] Multi-agent support baseline - roster/scoping/runtime routing foundation complete; polish phases remain active in current focus Notes --- - The old predictor-centered roadmap is retired as a public framing. Learned ranking can remain experimental, but Signet's product claim is portable agent state with inspectable extraction, provenance, and context selection. - This roadmap is directional. [Spec Index](./docs/specs/INDEX.md) remains the execution contract and dependency source of truth.