Engram Go

Every time you close your AI coding session, it forgets everything. The JWT library you chose. The expiry bug you spent an afternoon on. The pattern you explicitly rejected. Gone. Next session, the agent starts from zero and you start explaining.

# Session start — before touching any code
memory_recall("session handoff recent decisions", project="myapp")

# After settling on a technical choice
memory_store(
    "Chose RS256 over HS256: the API gateway needs to verify tokens without
     holding the signing secret. HS256 would require distributing the key to
     every service. Do not change this without updating the gateway config.",
    memory_type="decision",
    project="myapp"
)

Local-First by Default

Your memories stay on your machine.

Engram stores everything locally by design. Your PostgreSQL keeps every memory. Embeddings run locally via Ollama in local-only mode. Nothing leaves your machine unless you explicitly send it. Supported bootstrap modes:

• Local-only (recommended for fresh clones) — no external dependencies; start with docker-compose.local.yml.
• Hybrid — external Ollama/Olla router, started by make up via docker-compose.yml. Requires ENGRAM_ROUTER_URL in .env (legacy LITELLM_URL fallback accepted, retained for backward compatibility; see issue #634). PostgreSQL defaults to the bundled postgres service; set POSTGRES_HOST / POSTGRES_PORT only when you are pointing Engram at an external database.

Both setups share the same PostgreSQL backend, API contract, and tool set. Swap profiles without data loss or schema migration.

# Fresh-clone default: Local-only (zero external dependencies)
make init
docker volume create engram_pgdata
docker volume create ollama_ollama_storage
make build-postgres
docker compose -f docker-compose.local.yml up -d ollama
docker compose -f docker-compose.local.yml exec ollama ollama pull bge-m3
docker compose -f docker-compose.local.yml up -d
go run ./cmd/engram-setup --url http://127.0.0.1:8788

# Default home-network endpoint
# Hybrid profile checklist before `make up`:
# - `POSTGRES_PASSWORD` and `ENGRAM_API_KEY` are generated by `make init`
# - `ENGRAM_ROUTER_URL` (or legacy `LITELLM_URL`) points at your external router
# - `POSTGRES_HOST` / `POSTGRES_PORT` may be omitted for the bundled `postgres` service;
#   set them only for an external PostgreSQL server
make init
make up
make setup

What makes this different

Finds what you mean, not just what you typed. BM25 keyword search and 1024-dimensional semantic vectors run simultaneously. Searching "database lock timeout" finds your note about "WAL mode contention under load" — no shared words, close meaning. When Ollama is unavailable, search degrades gracefully to BM25+recency. Your results never disappear because an external service went down.

Weights by recency automatically. Exponential decay at 1% per hour. Yesterday's decision outranks one from six months ago. Nothing is deleted; old memories step back. Six-month-old memories are still there if nothing more recent matches.

Surfaces connected memories without being asked. A knowledge graph links decisions to the bugs they caused and the patterns they require. Recall one; get its neighborhood. Store a bug report, store the architectural pattern that caused it, connect them with a causes edge. Now any query about the pattern automatically surfaces the bug — you don't have to remember to ask for both.

Stores documents, not just notes. memory_store_document handles up to 500,000 characters. Engram chunks at sentence boundaries and embeds each chunk independently. A 20,000-word architecture document is searchable at the paragraph level — a query about authentication surfaces the auth section, not the whole document.

Quick Start

Prerequisites

• Docker Engine 20.10+ and Docker Compose 2.0+ — check with docker --version and docker compose version
• Go 1.26.3 or newer (matches go.mod) — check with go version; download from https://go.dev/dl/
• 4 GB RAM free — Ollama keeps the embedding model in memory
• 2 GB disk — PostgreSQL volume + Ollama model download (both cached on restart)

Initialize and Start

git clone https://github.com/petersimmons1972/engram-go.git && cd engram-go
make init

# Fresh-clone first run (recommended): local-only / no external backend
make build-postgres
docker compose -f docker-compose.local.yml up -d ollama
docker compose -f docker-compose.local.yml exec ollama ollama pull bge-m3
docker compose -f docker-compose.local.yml up -d
go run ./cmd/engram-setup --url http://127.0.0.1:8788

# Default home-network endpoint (requires operator environment)
# Hybrid `.env`: set `ENGRAM_ROUTER_URL` (or legacy `LITELLM_URL`) for your
# external router. `POSTGRES_HOST` / `POSTGRES_PORT` default to `postgres:5432`
# and only need overrides for an external PostgreSQL server.
make init
make up
make setup

Local Docker profiles listen at http://localhost:8788. make setup defaults to http://localhost:8788; pass --url to point at a remote host when you want Claude Code connected to an external server. Cold start: ~200ms. Idle memory: 18 MB.

Environment Configuration

If you prefer to author .env manually rather than using make init, see .env.example for all available variables, defaults, and descriptions. Key variables:

Connect to Claude Code

After setup:

/mcp

All 18 visible tools (+ 28 hidden maintenance tools, callable via tools/call) activate immediately. Four optional AI-enhanced tools (memory_ask, memory_reason, memory_explore, memory_query_document) activate when ANTHROPIC_API_KEY is set in .env. memory_diagnose is a built-in hidden tool, not AI-gated.

Setup Target and Config Writes

make setup defaults to http://localhost:8788; pass --url to point at a remote host. For local-only Docker, run the setup tool directly:

go run ./cmd/engram-setup --url http://127.0.0.1:8788

The setup tool health-checks the effective server URL, calls /setup-token, and writes the bearer-token MCP entry for Claude Code. Dry-run output redacts the token before printing. If /setup-token is unavailable during bootstrap, it validates fallback credentials from ~/.config/engram/api_key first and ~/projects/engram-go/.env (ENGRAM_API_KEY) second before writing anything.

Config files touched:

• ~/.claude/mcp_servers.json — created or updated with mcpServers.engram.
• ~/.claude.json — updated only when the file already has a mcpServers block; otherwise skipped.

Use make setup-dry-run or go run ./cmd/engram-setup --dry-run to print the effective server, endpoint, redacted entry, and target plan without writing files.

Important: RFC1918 and /setup-token

When Engram is in Docker (both profiles), it automatically accepts /setup-token requests from the Docker bridge (RFC1918). If you run Engram outside Docker and need to call /setup-token from a LAN address, add to .env:

ENGRAM_SETUP_TOKEN_ALLOW_RFC1918=1

Without this, /setup-token only accepts loopback (127.0.0.1 / ::1). See Operations → HTTP Endpoints for the full endpoint contract.

Configuration & Deployment

Read-Only Tool Permissions (Claude Code)

Engram's strictly read-only tools (memory_fetch, memory_list, memory_status, memory_history, memory_timeline, memory_projects, audit/episode listings, constraint checks, memory_diagnose) carry the MCP ReadOnlyHint: true annotation. memory_recall and memory_query are excluded because callers can opt into record_event=true, which stores retrieval telemetry. Claude Code's plan mode uses that annotation for permission gating.

Mutating tools (memory_store, memory_correct, memory_forget, memory_consolidate, memory_delete_project, etc.) intentionally prompt for permission on first use. To allow all engram tools without further prompts, add to ~/.claude/settings.json:

{
  "permissions": {
    "allow": ["mcp__engram__*"]
  }
}

For a narrower allowlist, copy the permissions.allow snippet logged by engram at startup — it lists only the read-only tool names.

Architecture

Your AI client speaks MCP over SSE. Engram registers 46 tools by default (18 visible in tools/list + 28 hidden but callable via tools/call), all running entirely locally. Four AI-enhanced tools (4) activate when ANTHROPIC_API_KEY is set, bringing the total to 50. PostgreSQL with pgvector stores everything. Ollama (local) runs the embeddings.

New in v3

RAG Queries: memory_ask

Ask a natural-language question against your stored memories. Returns a synthesized answer with citations — not a list of chunks, a direct response.

memory_ask(
    question="What did we decide about authentication and why?",
    project="myapp"
)
# → "You chose RS256 JWT with 24h expiry stored in httpOnly cookies.
#    The decision was driven by the need to verify tokens in the API gateway
#    without distributing the signing secret. localStorage was explicitly
#    rejected due to XSS risk. (memories: auth-decision-001, security-pattern-003)"

Document Storage: memory_store_document

Store up to 500,000 characters — architecture documents, meeting transcripts, entire codebases. Engram chunks at sentence boundaries and makes every paragraph individually searchable.

memory_store_document(
    content=entire_architecture_document,  # up to 500k chars
    memory_type="architecture",
    project="myapp"
)
# Later: search surfaces the specific section, not the whole document
memory_query_document(doc_id=doc_id, query="authentication flow")

Aggregate Queries

Query the shape of your memory store without reading individual memories.

memory_aggregate(by="memory_type", project="myapp")
# → [{label: "decision", count: 47}, {label: "error", count: 12}, ...]

memory_aggregate(by="failure_class")
# → [{label: "vocabulary_mismatch", count: 8}, {label: "stale_ranking", count: 3}, ...]

Retrieval Miss Tracking

When memory_recall returns nothing useful, log the miss with a failure class. This feeds the retrieval quality benchmark and makes future recall better.

event_id only appears in memory_recall's response when the call passes record_event=true (off by default so plain recall stays side-effect free). Pass it through when you plan to follow up with memory_feedback:

memory_recall(query="deployment runbook", project="myapp", record_event=True)
# → {..., event_id: "0197f3c1-...", feedback_hint: "Call memory_feedback with this event_id and the memory_ids you used"}

memory_feedback(
    event_id="<event_id from recall>",
    memory_ids=[],
    failure_class="vocabulary_mismatch"  # or: aggregation_failure, stale_ranking,
                                          #     missing_content, scope_mismatch, other
)

New in v3.1

Decay Audit System

Track retrieval drift over time with canonical query snapshots. Register a set of reference queries, run them on a schedule, and measure how results shift between runs using RBO (rank-biased overlap) and Jaccard similarity. Alerts when drift exceeds threshold.

# Register a reference query
memory_audit_add_query(project="myapp", query="deployment procedures", description="CI/CD runbook recall")

# Take a snapshot and see drift vs previous run
memory_audit_run(project="myapp")
# → [{query: "deployment procedures", rbo_vs_prev: 0.94, additions_count: 1, removals_count: 0, alert: false}]

# Browse snapshot history for a query
memory_audit_compare(query_id="cq-abc123", limit=10)

Five new tools: memory_audit_add_query, memory_audit_list_queries, memory_audit_deactivate_query, memory_audit_run, memory_audit_compare.

Adaptive Weight Tuning

Failure-class events now feed a background tuner that adjusts the four search weights per project. Dominant vocabulary_mismatch events shift weight toward BM25; dominant stale_ranking shifts toward recency. Adjustments fire at most once per 7 days after ≥ 50 failure events, within hard per-weight guardrails.

memory_weight_history(project="myapp")
# → {current_weights: {vector: 0.40, bm25: 0.35, recency: 0.10, precision: 0.15},
#    history: [{applied_at: "...", weights: {...}, trigger_data: "..."}]}

Expanded Relation Types

Knowledge graph now supports 11 typed edges (up from 7):

Pluggable Embedder Interface + Eval Tool

Inspect the embedding models available through Ollama before storing memories.

# See what models are installed and recommended
memory_models()

other Failure Class

memory_feedback now accepts failure_class="other" for misses that do not fit the four specific categories. Tracked separately in memory_aggregate(by="failure_class").

Diagnostics

scripts/verify-recall-fetch.sh surfaces recall→fetch orphans: cases where memory_recall returns handle IDs that memory_fetch then reports as not found. These indicate index entries whose backing store rows are missing. The script calls /quick-recall for a configurable query, issues a memory_fetch for every returned ID, and prints a summary of total_handles / fetch_success / fetch_404 / fetch_other_error. It exits non-zero when any orphans are found, making it safe to run in CI or from a cron job. Run ./scripts/verify-recall-fetch.sh --help for all options.

Documentation

v3.0 vs v2 vs v1

v1 was Python. v2 rewrote in Go. v3.0 adds required authentication, auto-episode starts on every SSE connection, 35 tools, document storage, RAG queries, and aggregate analysis. v3.1 adds decay audit, adaptive weight tuning, expanded relation types, embedder eval, and 8 new tools.

Credits

v3.1 features were developed in dialogue with open-brain-template by Myles Bryning. The decay audit concept, supports / derived_from / part_of / follows relation types, and pluggable embedder registry are derived from open-brain-template's architecture, adapted to engram-go's local-only, Ollama-first constraint. The BM25+vector+recency blend, failure-class taxonomy, and knowledge-graph-based retrieval originated in engram-go and were independently incorporated by open-brain-template.

GPL v3 — see LICENSE