instantDocumentation / single-node · source MVP
Documentation
Install the service, save your first record and connect an agent. The guides cover checkpoints, worker permissions and running your own server.
Start here
Quick startInstall instantKV, store your first memory and restore a checkpoint.Connect through MCPFive memory tools, seven KV/checkpoint tools and checkpoint instructions.OpenCode native memoryInstall local MCP memory for OpenCode and run an explicit save, restart and recall demonstration.OpenCode memory controllerConnect the lifecycle controller to OpenCode with your existing model provider.Memory API & local modelsRemember, recall, search, browse and forget. Find memories by topic, tag, time or keywords. Try the local-model example.
Use the service
Memory & compactionWhat to save before compaction and how to load it afterward.Memory lifecycle controllerOptional local integration for reviewed facts, corrections, profiles and forgetting.Local agents & swarmsGive workers shared project facts and separate private notes.CLI referenceCommands for structured memories, checkpoints, setup and health.HTTP referenceAuthenticated routes, conditional writes, TTL, errors and examples.
Run a node
Evidence
Performance & device targetsFull retrieval results, Mac measurements and planned phone tests.Full retrieval reportAll 500 LongMemEval-S questions and all ten LoCoMo histories.Benchmark methodologyRecorded workloads, hardware, load conditions and reproduction commands.Evaluation policyFrozen tests, reproducible results and honest comparison limits.Memory benchmark notesFull memory suites, retrieval wins, remaining gaps and raw evidence.
Build with us
Architecture & schemaRust, the bounded engine, redb transactions and storage schema.Technology & learning mapUnderstand the service, memory lifecycle, agent runtime and website stack.Repository maintenanceFile ownership, canonical guides, review rules and verification.ContributingSource map, local checks, invariants and the contribution workflow.SecurityReport vulnerabilities privately and understand deployment boundaries.Website & deploymentBuild, verify and deploy this static Cloudflare site.Search & agent discoveryCanonical pages, Markdown exports, crawl checks and evidence-based discoverability.Private product measurementTrack release downloads and recent GitHub traffic in a private ledger. Counting and privacy limits.Engineering referencesPractices adopted from database, low-level and KDE projects.
Project
Cleanup & release planOrdered work, release gates and validation requirements.RoadmapSource MVP status, local-model evaluation, phone integration and optional retrieval plans.Local AI memory: when instantKV fitsWho it helps, what makes it useful and when another memory tool fits better.FeaturesAvailable features, the source MVP and planned work.ChangelogChanges by release and work awaiting the next release.