memmesh-go

Official Go client for MemMesh — memory + prediction for AI agents. Semantic recall, a bi-temporal knowledge graph, belief revision, reflection, and calibrated forecasting behind an idiomatic Go client.

go get github.com/ThinkfleetAI/memmesh-go

Quickstart

package main

import (
	"context"
	"fmt"

	memmesh "github.com/ThinkfleetAI/memmesh-go"
)

func main() {
	ctx := context.Background()
	mm := memmesh.New("sk-...", "proj_...") // apiKey, projectId

	// Remember something
	mm.Memory.Observe(ctx, memmesh.Observe{
		Subject: memmesh.Subject{Kind: "contact", ExternalID: "sarah"},
		Content: "Prefers email over phone.",
	})

	// Recall it, semantically
	hits, _ := mm.Memory.Search(ctx, "how to reach sarah", memmesh.SearchOpts{Limit: 5})
	for _, h := range hits {
		fmt.Println(h.Content)
	}

	// Synthesize higher-order insights, with provenance
	res, _ := mm.Memory.Reflect(ctx, memmesh.ReflectOpts{MaxInsights: 3})
	for _, in := range res.Insights {
		fmt.Printf("%s  (%.0f%%)\n", in.Content, in.Confidence*100)
	}

	// Point-in-time knowledge graph — "what did we believe on date X"
	edges, _ := mm.Context.QueryGraph(ctx, memmesh.GraphOpts{AsOf: "2026-03-01T00:00:00Z"})
	_ = edges
}

Surface

Service	Highlights
mm.Memory	Observe, Create, Search, List, Update, Delete, Stats, Confirm, Promote, Feedback, Reflect, PrefetchRelated, Dedup, BackfillEmbeddings
mm.Lattice	Predict (any target, calibrated + abstaining), Mine, Profile, PredictByCohort, Calibration
mm.Context	Build, BatchBuild (bulk, ≤500), QueryGraph (point-in-time)
mm.Events · mm.Alerts	Emit/Poll events; alert rules + fires
mm.Learning	RecordDecision → RecordOutcome → Effectiveness (the self-improving loop)
mm.Typed · mm.Compliance · mm.Health	Typed attributes; export/erase/audit/packs; healthcare pack

Config

• memmesh.New(apiKey, projectID, opts...)
• memmesh.WithBaseURL("https://your-host") — self-hosted or a different region
• memmesh.WithHTTPClient(&http.Client{...}) — timeouts, proxies, retries

Errors from the API are *memmesh.APIError (.Status, .Code).

Apache-2.0 · memmesh.ai · docs