|
9 | 9 | # LLM key, no embedding key, no API auth, and no scheduled eviction. Maintenance |
10 | 10 | # sweeps that only bound internal metadata may be enabled unless marked otherwise. |
11 | 11 | # |
12 | | -# Run `npx @agentmemory/agentmemory init` to copy this file into place |
13 | | -# automatically. Run `npx @agentmemory/agentmemory doctor` to verify that |
14 | | -# the daemon reads the env you expect. |
| 12 | +# Run `npx -y @agentmemory/agentmemory@latest init` to copy this file into |
| 13 | +# place automatically. Run `npx -y @agentmemory/agentmemory@latest doctor` |
| 14 | +# to verify that the daemon reads the env you expect. |
15 | 15 | # |
16 | 16 | # Defaults shown in comments. Listed in priority order — the first key |
17 | 17 | # present wins on the LLM detection path (see src/config.ts::detectProvider). |
|
21 | 21 | # ----------------------------------------------------------------------------- |
22 | 22 | # |
23 | 23 | # Without a provider key, agentmemory runs in noop mode: observations are |
24 | | -# indexed via zero-LLM synthetic compression, hybrid search still works, |
| 24 | +# indexed via zero-LLM synthetic compression and BM25 recall still works, |
25 | 25 | # but LLM-backed summarisation / reflection / consolidation are disabled. |
| 26 | +# A provider key alone does not enable LLM-written observation compression; |
| 27 | +# that path also requires AGENTMEMORY_AUTO_COMPRESS=true. |
26 | 28 | # The detection order is OPENAI_API_KEY → MINIMAX_API_KEY → ANTHROPIC_API_KEY |
27 | 29 | # → GEMINI_API_KEY → OPENROUTER_API_KEY → noop. |
28 | 30 |
|
|
64 | 66 | # 2. Embedding provider — auto-detected, override via EMBEDDING_PROVIDER |
65 | 67 | # ----------------------------------------------------------------------------- |
66 | 68 | # |
67 | | -# Without an embedding key, agentmemory runs in BM25-only mode for hybrid |
68 | | -# search. Detection order: EMBEDDING_PROVIDER override → GEMINI_API_KEY → |
| 69 | +# Without an embedding provider, vector embeddings are disabled. `mem::search` |
| 70 | +# uses BM25; `mem::smart-search` can also fuse structural graph matches when |
| 71 | +# graph data exists. Remote |
| 72 | +# provider detection order: EMBEDDING_PROVIDER override → GEMINI_API_KEY → |
69 | 73 | # OPENAI_API_KEY → VOYAGE_API_KEY → COHERE_API_KEY → OPENROUTER_API_KEY → |
70 | | -# local (Xenova/all-MiniLM-L6-v2, 384-dim). |
| 74 | +# BM25-only. Local embeddings are an explicit opt-in, not the keyless default. |
| 75 | +# `EMBEDDING_PROVIDER=local` uses Xenova/all-MiniLM-L6-v2 (384-dim) on-device; |
| 76 | +# the first embedding request downloads the model and needs network access. |
71 | 77 |
|
72 | 78 | # EMBEDDING_PROVIDER=local # local | openai | voyage | cohere | gemini | openrouter |
73 | 79 |
|
|
107 | 113 | # 5. Behaviour flags |
108 | 114 | # ----------------------------------------------------------------------------- |
109 | 115 |
|
110 | | -# AGENTMEMORY_AUTO_COMPRESS=true # Run LLM compression on every observation batch (requires a provider key). Default off — synthetic compression handles most cases. |
| 116 | +# AGENTMEMORY_AUTO_COMPRESS=true # Run LLM-written compression on every observation batch. Requires BOTH this flag and a configured provider; a provider alone leaves synthetic compression active. |
111 | 117 | # AGENTMEMORY_INJECT_CONTEXT=true # Inject recalled memories back into agent prompts (#143). Default off — hooks capture observations but do not modify conversation. |
112 | 118 | # CONSOLIDATION_ENABLED=false # Runs by default with an LLM provider; set false to opt out of the 4-tier pipeline. |
113 | 119 | # CONSOLIDATION_DECAY_DAYS=30 # Age (days) after which non-reinforced memories decay during consolidation |
|
150 | 156 | # AGENTMEMORY_PROBE_TIMEOUT_MS=2000 # MCP shim livez probe timeout |
151 | 157 | # AGENTMEMORY_URL=http://localhost:3111 # REST base URL — honored by status, doctor, MCP shim |
152 | 158 | # AGENTMEMORY_VIEWER_URL=http://localhost:3113 # Override the viewer URL printed by `agentmemory status` |
| 159 | +# Persistent iii state; same override as `--data-dir`, used by native and |
| 160 | +# Docker starts. Explicit flag/env wins. Otherwise instance 0 reuses an |
| 161 | +# existing ./data/state_store.db or ./data/iii-config.yaml before defaults: macOS |
| 162 | +# ~/Library/Application Support/agentmemory; Linux $XDG_DATA_HOME/agentmemory |
| 163 | +# or ~/.local/share/agentmemory; Windows %APPDATA%\agentmemory. |
| 164 | +# AGENTMEMORY_DATA_DIR=/absolute/path |
| 165 | +# AGENTMEMORY_USE_DOCKER=1 # Prefer bundled Docker even when a compatible native iii binary exists |
153 | 166 | # AGENTMEMORY_EXPORT_ROOT=~/agentmemory-backup # Default destination for `agentmemory export` |
154 | 167 |
|
155 | 168 | # STANDALONE_MCP=1 # MCP shim only — bypass the worker and run @agentmemory/mcp in-process |
|
169 | 182 | # 7. Ports |
170 | 183 | # ----------------------------------------------------------------------------- |
171 | 184 |
|
172 | | -# III_REST_PORT=3111 # REST API port (also affects viewer at +2) |
173 | | -# III_STREAMS_PORT=3112 # Streams API port |
174 | | -# III_ENGINE_URL=ws://localhost:49134 # iii-engine WebSocket URL (used by the worker) |
| 185 | +# III_REST_PORT=3111 # REST API + MCP HTTP anchor |
| 186 | +# III_STREAM_PORT=3112 # Preferred streams override; defaults to REST + 1 |
| 187 | +# III_STREAMS_PORT=3112 # Legacy alias for III_STREAM_PORT |
| 188 | +# III_VIEWER_PORT=3113 # Viewer bind port; defaults to REST + 2 |
| 189 | +# III_ENGINE_PORT=49134 # Engine bind port; defaults to REST + 46023 |
| 190 | +# III_ENGINE_URL=ws://localhost:49134 # Explicit/remote worker WebSocket URL; takes precedence over derived engine port |
| 191 | +# Derivation applies only while the corresponding explicit port/URL is unset. |
| 192 | +# `--instance 1` selects 3211/3212/3213/49234 and an instance-1 data subdir. |
175 | 193 |
|
176 | 194 | # ----------------------------------------------------------------------------- |
177 | 195 | # 8. iii engine pin |
|
0 commit comments