List view
End-to-end lifecycle simulation test suite — validates the full mnemonic user journey from fresh install through 3 months of usage, exercising all 8 cognitive agents, encoding pipeline, consolidation, dreaming, and retrieval quality under realistic scale.
No due date•12/12 issues closedReduce noise in recall, prioritize high-signal memories, improve encoding consistency, and make the pipeline observable. Addresses friction points identified by Claude after the Claude-First Memory milestone shipped.
No due date•8/8 issues closedMake Mnemonic a collaborative knowledge system for its primary consumer (Claude Code via MCP). Closes dead feedback loops, enables memory correction, surfaces the association graph, adds recall transparency, and makes the system genuinely learn from interactions. Builds on Milestone 6 which addressed surface-level friction.
No due date•9/9 issues closedMake mnemonic actually useful for its primary consumer (Claude Code via MCP). Addresses duplicate recall results, missing filters, shallow feedback loops, and encoding dedup.
No due date•6/6 issues closedRemediate compounding gaps and hardcoded tunables identified by codebase audit. 7 issues across 3 phases: (1) close retrieval feedback loops, (2) externalize ~40 hardcoded agent thresholds to config.yaml, (3) batch embedding, self-sustaining patterns, dreaming quality gates, and test determinism.
No due date•9/9 issues closedFelix-LM v3 pretraining: HP optimization, 100M encoder training, GGUF export verification
No due date•2/2 issues closedDownload, filter, clean, tokenize, and mix domain-specific pretraining data for Felix-LM v3. Custom tokenizer, 10B tokens across neuroscience papers, code, FineWeb-Edu, StackOverflow, structured data, and git commits.
No due date•7/7 issues closedSystematic cleanup of duplicated code, dead code removal, and modularization across the Go backend and web dashboard. Identified via comprehensive code review. ## Scope - Extract shared utility functions from agents into `internal/agent/agentutil/` - Remove dead/duplicate type definitions - Consolidate duplicated constants and helpers across API, store, config - Extract shared test infrastructure - Dashboard JS deduplication ## Guiding Principles - Pure refactoring — no behavior changes - Each PR should pass `make check && make test` - Small, focused PRs (one finding per PR)
No due date•10/10 issues closedEverything needed to make Mnemonic ready for public use. Tracked from the March 2026 readiness audit. ## Tiers - **Tier 1 (Blockers)**: Must fix before any public announcement - **Tier 2 (High Impact)**: Should fix for a solid first impression - **Tier 3 (Nice to Have)**: Post-launch improvements ## Target State A developer can `brew install mnemonic`, run `mnemonic serve`, get clear feedback if LM Studio isn't running, and have confidence their data is safe.
Due by May 31, 2026•16/16 issues closed