Contributors
Memix is all of us
Our contributors 1
Thank you for supporting Memix.
About
Memix isn't just another AI coding assistant. It's an autonomous intelligence layer that runs silently in the background of your IDE, continuously observing, predicting, and remembering everything about your codebase.
While other tools wait for you to ask a question, Memix is already understanding the context, mapping your architecture, and preparing the exact codebase memories your AI needs before you even open the chat.
Memix Features
✅ Persistent Project Brain (Redis-backed)
Stores identity, session state, tasks, decisions, patterns, file map, and known issues so your AI resumes with continuity.
✅ Daemon-Managed JSON Mirror (.memix/brain/*.json)
Every successful brain write is mirrored to workspace-local JSON files. This gives your AI and tooling an instant, file-based read path without direct Redis access.
✅ Safe AI Writeback via pending.json
AI agents can propose writes by creating .memix/brain/pending.json. The daemon validates, applies, writes pending.ack.json, and clears pending input.
✅ AST-driven Code DNA + Architecture Explainability
Memix now computes project DNA from Tree-Sitter ASTs instead of keyword heuristics. That means language-aware cyclomatic complexity, structural pattern tagging, exported symbol detection, architecture inference, hot-zone ranking, circular risk detection, and a short explainability summary you can inject directly into AI context.
✅ Configurable DNA Rules Per Project
Observer DNA can be tuned with workspace-local overrides in .memix/dna_rules.toml, letting you classify custom folders, tag proprietary patterns, and handle project-specific edge cases without changing the daemon code. Start from docs/dna_rules.toml.example and copy it into your workspace as .memix/dna_rules.toml.
✅ Autonomous Codebase Observation
Memix quietly monitors your active workspace. As you scaffold new files, refactor architecture, or fix bugs, the Memory Engine natively understands the intent behind your changes. It maps how your functions connect together in real-time.
✅ Zero-Latency Context Routing
Tired of manually highlighting code so your AI understands what you're talking about? Memix pre-loads the exact files, dependencies, and historical decisions into your AI's context window. When you ask a question, the AI already knows the answer.
✅ Ironclad Privacy & Offline Execution
Your code never leaves your machine unless you explicitly prompt your AI model. Memix operates entirely natively on your hardware. It builds lightning-fast semantic graphs, vector indexes, and structural models running at purely native performance.
✅ Seamless Team Sync
Memix ensures your entire developer team shares the exact same codebase "brain." Architectural decisions, API boundaries, and feature plans sync effortlessly without conflict, ensuring every developer—and every AI agent—is on the identical page.
✅ One-click Health Check
Validates required brain keys, detects invalid shapes, staleness, and oversized entries.
✅ Brain Key Coverage (Advanced)
Shows which brain keys exist, their sizes, and their taxonomy so you can quickly spot missing categories before an AI session.
✅ Prompt Pack Preview + One-click Copy (Advanced)
Generates a ready-to-paste context bundle (identity, session state, patterns, decisions, known issues, tasks, file map). Includes a token estimate so you can fit it into any model’s context window.
✅ Mirror Import / Export + Migrations
- Daemon endpoints for full mirror export/import.
- Schema migrations endpoint for project backfills (for example: vector backfill).
- Helps keep old projects up to date as Memix capabilities evolve.
✅ Token utilities (daemon)
Exact token counting and budget-based context selection.
✅ Observer + Session APIs
- /api/v1/observer/graph
- /api/v1/observer/changes
- /api/v1/observer/dna
- /api/v1/observer/dna/otel
- /api/v1/session/current
- /api/v1/session/replay
✅ Local daemon over Unix Socket + TCP
A local Axum (Rust) daemon powers memory APIs, rules generation, observer snapshots, migrations, and mirror sync.
✅ Smarter Similarity Search
Hybrid similarity now combines normalized vector similarity (cosine) with keyword overlap for stronger relevance.
✅ Optional Real Embeddings (feature-flagged)
all-MiniLM-L6-v2 via fastembed is available behind real_embeddings so default builds remain stable on environments without ONNX runtime support.
Hard advantages — why Memix improves AI-chat workflows ?
- Prompt Pack (copy/paste ready) A curated, structured bundle you can paste into any AI chat to eliminate “re-explain the repo” prompts.
- Token-aware context Memix can estimate tokens for your Prompt Pack so you can stay under model limits.
- Decision guardrails Persist architectural decisions and feed them forward so the AI stops re-debating solved choices.
- Task-aware prompting Session state + tasks keep the AI aligned with what you’re actually doing right now.
- Mirror-first reliability Workspace-local JSON mirror ensures low-latency reads and resilient workflows even when network/Redis conditions fluctuate.
- Migration-safe evolution As memory schema and embedding capabilities improve, migration hooks keep existing projects consistent instead of drifting.
- Architecture-aware AI context Code DNA summaries give your AI a compact briefing of architecture style, hot zones, dependency depth, rule-derived patterns, and circular dependency risks before it starts making edits.
Why Memix matters?
Memix is not only about remembering facts. It is designed to make AI-assisted coding materially better across the full development loop.
- Faster coding Memix reduces the time you spend re-explaining your project to AI. The daemon observes structure, recent changes, architectural decisions, and active work areas so the AI starts closer to the real context.
- Better code quality By preserving project decisions, patterns, file roles, dependency relationships, and Code DNA summaries, Memix helps the AI produce changes that fit the codebase instead of generic answers that fight your architecture.
- Less context loss In long sessions, after IDE restarts, or across multiple contributors, Memix keeps continuity through persistent brain state, task tracking, project mirrors, and observer snapshots.
- Safer architectural changes Memix surfaces hot zones, dependency depth, circular risks, and active development areas. That gives the AI and the developer better visibility into where edits are risky and where changes are likely to ripple.
- More useful than memory alone Traditional memory tools mostly store notes. Memix also watches the codebase itself, builds structural intelligence from ASTs, exposes observer APIs, and prepares compact explainability summaries that can be injected directly into AI workflows.
- Higher leverage for teams Memix helps multiple developers and AI agents work from the same shared understanding of architecture, conventions, tasks, and known issues instead of rebuilding that understanding from scratch every session.
In short: Memix helps you code faster, keep quality higher, preserve architectural consistency, and make AI assistance feel less stateless.
Our team
Soufiane Loud...
Admin