hermes - 💡(How to fix) Fix [Bug] Holographic memory context injection never fires — get_relevant_memories not called before turns (v0.14.0)

Environment

• Hermes version: v0.14.0 (2026.5.16)
• OS: Windows 10 + WSL2 (Ubuntu 24)
• HERMES_HOME: /mnt/e/ai/hermes_agent/brains/ (non-default, external SSD)
• Launch command: hermes --tui
• Memory provider: holographic
• Models tested: qwen/qwen-2.5-72b-instruct (OpenRouter), MiniMax-M2.7, google/gemini-2.0-flash-exp
• numpy: installed (2.4.6)

Summary

Holographic memory registers and activates correctly at session start, and auto_extract correctly stores facts at session end. However, the context injection hook (get_relevant_memories / pre_llm_call) never fires before turns — facts are never surfaced to the model, causing hallucination on recall.

Expected behavior

Per the docs, when a memory provider is active Hermes should:

1. Prefetch relevant memories before each turn (background, non-blocking)
2. Inject provider context into the system prompt

Actual behavior

• hermes memory status correctly shows holographic as active
• Session init log shows: Memory provider 'holographic' registered (2 tools)
• Facts are correctly stored in SQLite (auto_extract: true works after numpy installed)
• No log entries for prefetch, inject, relevant, or holographic appear during conversation turns
• Model receives large context (~19,000 tokens, confirming tools are sent) but hallucinates on recall — stored facts never appear in context
• tool_turns=0 on every turn across all models tested

What IS working

• hermes memory status shows holographic active ✓
• Memory provider 'holographic' registered (2 tools) appears at session start ✓
• Auto-extracted N facts from conversation fires at session end ✓
• Facts correctly written to SQLite DB ✓
• /tools lists tools correctly when launched with hermes --tui ✓

What is NOT working

• No prefetch or context injection log entries during any conversation turn ✗
• Stored facts never surfaced to model ✗
• Model hallucinates instead of recalling stored facts ✗

Relevant logs

Session start (working correctly):

INFO agent.memory_manager: Memory provider 'holographic' registered (2 tools) INFO run_agent: Memory provider 'holographic' activated

During conversation turns (injection never appears):

INFO agent.conversation_loop: conversation turn: ... msg='do you remember my favorite color?' INFO agent.conversation_loop: API call #1: model=qwen/qwen-2.5-72b-instruct in=19101 out=18 INFO agent.conversation_loop: Turn ended: reason=text_response tool_turns=0

Session end (auto_extract working):

INFO plugins.memory.holographic: Auto-extracted 1 facts from conversation

Fact store contents (correctly stored):

4|actually, my favorite color is gunmetal grey|user_pref||0.5|0|0|2026-05-23 3|nope, my favorite color is gunmetal grey|user_pref||0.5|0|0|2026-05-23

Additional notes

• Discovered that launching with hermes -tui (single dash) silently sets toolsets to empty, causing a completely different set of symptoms. Correct launch is hermes --tui.
• numpy was missing initially, causing auto_extract to fail silently — installing numpy fixed storage but not retrieval.
• Issue persists regardless of model or provider used.
• Related issues: #4781, #22907

hermes doctor output

◆ Security Advisories ✓ No active security advisories

◆ Python Environment ✓ Python 3.11.14 ✓ Virtual environment active

◆ Required Packages ✓ OpenAI SDK ✓ Rich (terminal UI) ✓ python-dotenv ✓ PyYAML ✓ HTTPX ✓ Croniter (cron expressions) (optional) ✓ python-telegram-bot (optional) ✓ discord.py (optional)

◆ Configuration Files ✓ /mnt/e/ai/hermes_agent/brains/.env file exists ✓ API key or custom endpoint configured ✓ /mnt/e/ai/hermes_agent/brains/config.yaml exists ✓ Config version up to date (v23)

◆ xAI Model Retirement (May 15, 2026) ✓ No retired xAI models in config

◆ Auth Providers ⚠ Nous Portal auth (not logged in) ⚠ OpenAI Codex auth (not logged in) → No Codex credentials stored. Run hermes auth to authenticate. → codex CLI not installed (optional — only required to import tokens from an existing Codex CLI login) ⚠ Google Gemini OAuth (not logged in) ✓ MiniMax OAuth (logged in, region=global) ⚠ xAI OAuth (not logged in) → No xAI OAuth credentials stored. Select xAI Grok OAuth (SuperGrok Subscription) in hermes model.

◆ Directory Structure ✓ /mnt/e/ai/hermes_agent/brains directory exists ✓ /mnt/e/ai/hermes_agent/brains/cron/ exists ✓ /mnt/e/ai/hermes_agent/brains/sessions/ exists ✓ /mnt/e/ai/hermes_agent/brains/logs/ exists ✓ /mnt/e/ai/hermes_agent/brains/skills/ exists ✓ /mnt/e/ai/hermes_agent/brains/memories/ exists ✓ /mnt/e/ai/hermes_agent/brains/SOUL.md exists (persona configured) ✓ /mnt/e/ai/hermes_agent/brains/memories/ directory exists → MEMORY.md not created yet (will be created when the agent first writes a memory) → USER.md not created yet (will be created when the agent first writes a memory) ✓ /mnt/e/ai/hermes_agent/brains/state.db exists (44 sessions)

◆ Gateway Service ✓ Systemd linger enabled (gateway service survives logout)

◆ Command Installation ✓ Venv entry point exists (venv/bin/hermes) ✓ ~/.local/bin/hermes exists (non-symlink)

◆ External Tools ✓ git ✓ ripgrep (rg) (faster file search) ⚠ docker not found (optional) ✓ Node.js ✓ agent-browser (Node.js) (browser automation) ✓ Playwright Chromium (browser engine) ✓ Browser tools (agent-browser) deps (no known vulnerabilities)

◆ API Connectivity ✓ OpenRouter API

◆ Tool Availability ✓ browser ✓ clarify ✓ code_execution ✓ cronjob ✓ terminal ✓ delegation ✓ file ✓ memory ✓ moa ✓ session_search ✓ skills ✓ todo ✓ tts ✓ vision ✓ video ✓ kanban (runtime-gated; loaded only for dispatcher-spawned workers) ⚠ browser-cdp (system dependency not met) ⚠ computer_use (system dependency not met) ⚠ discord (missing DISCORD_BOT_TOKEN) ⚠ discord_admin (missing DISCORD_BOT_TOKEN) ⚠ feishu_doc (system dependency not met) ⚠ feishu_drive (system dependency not met) ⚠ homeassistant (system dependency not met) ⚠ image_gen (system dependency not met) ⚠ messaging (system dependency not met) ⚠ video_gen (system dependency not met) ⚠ web (missing EXA_API_KEY, PARALLEL_API_KEY, TAVILY_API_KEY, FIRECRAWL_API_KEY, FIRECRAWL_API_URL, FIRECRAWL_GATEWAY_URL, TOOL_GATEWAY_DOMAIN, TOOL_GATEWAY_SCHEME, TOOL_GATEWAY_USER_TOKEN) ⚠ x_search (missing XAI_API_KEY) ⚠ hermes-yuanbao (system dependency not met) ⚠ spotify (system dependency not met)

◆ Skills Hub ⚠ Skills Hub directory not initialized (run: hermes skills list) ⚠ No GITHUB_TOKEN (60 req/hr rate limit — set in /mnt/e/ai/hermes_agent/brains/.env for better rates)

◆ Memory Provider ✓ holographic provider active