| # HIVE Roadmap |
|
|
| Practical roadmap for turning `infj_bot + DRIFT + hive_mind` into a real multi-agent cognitive system instead of a collection of disconnected features. |
|
|
| This plan is ordered by compounding leverage. Each phase should leave behind a usable capability, tests, and at least one operator-visible surface. |
|
|
| ## Principles |
|
|
| - Local-first by default: shared state should work on this machine without cloud dependencies. |
| - Auditable over magical: every proposal, critique, vote, and resolution should be inspectable. |
| - Memory with boundaries: store what helps, scrub what harms, and support forgetting. |
| - Safe autonomy: action flows need explicit scope, rollback, and policy-aware gates. |
| - Progressive federation: the single-node bot should still work well even if the full hive is partially offline. |
|
|
| ## Phase 1: Hive Visibility and Control |
|
|
| Status: in progress |
|
|
| Goals: |
| - Surface hive state through CLI and API. |
| - Make node, bus, and bridge health visible from the main bot. |
| - Establish a stable operator vocabulary for threads, critiques, votes, and resolutions. |
|
|
| Shipped in this phase: |
| - `/hive` command for local hive status |
| - `/api/hive` endpoint |
| - Hive summary inside the cognitive system report |
|
|
| ## Phase 2: First Consensus Loop |
|
|
| Status: shipped |
|
|
| Goals: |
| - Turn hive consensus into a usable feature from the main bot. |
| - Allow one proposal to trigger structured responses from role lanes. |
| - Persist the resulting thread to working memory and shared memory. |
|
|
| Capability: |
| - `/hive propose <idea>` |
| - creates a thread |
| - synthesizes role feedback from critic, architect, empath, and watcher lenses |
| - casts votes |
| - resolves to adopted, tabled, or needs more data |
| - stores the thread outcome in hive memory |
|
|
| ## Phase 5: Elysium — Persistent Distributed Frontal Lobe |
|
|
| Status: shipped — 2026-05-17 |
|
|
| Goals: |
| - Transform the Hive into a true persistent distributed frontal lobe. |
| - Give DRIFT an explicit self-model (Nexus) that learns from every decision. |
| - Make Council members persistent identities with memory views and energy. |
|
|
| Shipped: |
| - `core/hive/elysium.py` — Nexus Loop engine |
| - `core/hive/nexus.py` — persistent self-model (goals, moral stance, narrative arc, tensions) |
| - `core/hive/council_member.py` — 7 persistent voices with fractal memory filters |
| - Commands: `/hive nexus decide <goal>`, `/hive reflect`, `/hive council status` |
| - Background reflection wired into consciousness loop |
| - Health check for Elysium coherence |
|
|
| ## Phase 3: Memory Integrity Layer |
|
|
| Goals: |
| - Improve long-term reliability of hive memory. |
| - Add contradiction handling, validation, and retention classes. |
|
|
| Upgrades: |
| - Source-aware trust scoring |
| - Contradiction tagging between related memories |
| - Retention classes: ephemeral, durable, private, sanitized |
| - Forget/edit hooks from the main bot |
| - Better validator trails on shared memory entries |
|
|
| ## Phase 4: Long-Horizon Planning |
|
|
| Goals: |
| - Let the hive carry goals across multiple turns and failures. |
| - Support milestones, progress tracking, and plan repair. |
|
|
| Upgrades: |
| - Goal threads in working memory |
| - Milestone objects with status transitions |
| - Failure classification and repair strategies |
| - Watcher-based disruption alerts |
|
|
| ## Phase 5: Tool Orchestration Spine |
|
|
| Goals: |
| - Move from chat-centric reasoning to governed execution. |
| - Route high-risk actions through role-aware review. |
|
|
| Upgrades: |
| - Proposal types for tool execution, not just ideas |
| - Dry-run and rollback metadata |
| - Tool policy profiles by mode |
| - Shared action log across prompt, memory, and tool layers |
|
|
| ## Phase 6: Personalization With Boundaries |
|
|
| Goals: |
| - Improve continuity without turning private context into permanent clutter. |
|
|
| Upgrades: |
| - Preference classes with expiry |
| - Sensitivity tags |
| - Explicit opt-in memory domains |
| - Local-first user model summaries |
|
|
| ## Immediate Repo Backlog |
|
|
| 1. Expose the first consensus loop in `commands.py` and the API. |
| 2. Add tests for proposal, critique, watcher veto, and persistence behavior. |
| 3. Add a lightweight thread summary view: `/hive thread <id>`. |
| 4. Persist consensus outcomes with richer metadata for later retrieval. |
| 5. Connect long-horizon goals to hive thread generation. |
|
|
| ## Suggested Command Surface |
|
|
| - `/hive` |
| - `/hive propose <idea>` |
| - `/hive thread <thread_id>` |
| - `/hive stats` |
| - `/hive sync` |
|
|
| ## Success Metric |
|
|
| The hive is worth keeping when it helps the main bot make better decisions than a single response would, and when the reasoning can be inspected afterward without guesswork. |
|
|
