my harness for niri
1
fork

Configure Feed

Select the types of activity you want to include in your feed.

at master 76 lines 3.6 kB view raw
1# ── Primary model (OpenAI-compatible) ──────────────────────────────────────── 2OPENAI_BASE_URL=https://api.openai.com/v1 3OPENAI_API_KEY= 4MODEL= 5 6# Used when the primary endpoint is unreachable or rate-limited (429/5xx). 7# Supports any OpenAI-compatible provider (OpenRouter, LM Studio, etc.). 8FALLBACK_OPENAI_BASE_URL=http://localhost:1234/v1 9FALLBACK_OPENAI_API_KEY= 10FALLBACK_MODEL= 11# Tool call policy for fallback requests: required, auto, none 12FALLBACK_TOOL_CHOICE=required 13# Optional headers 14FALLBACK_OPENAI_REFERER= 15FALLBACK_OPENAI_TITLE= 16# Context window enforcement for fallback. Defaults to true for localhost, false otherwise. 17FALLBACK_ENFORCE_CONTEXT_LIMIT= 18FALLBACK_N_CTX=4096 19FALLBACK_CONTEXT_MARGIN=256 20FALLBACK_HARD_OVERFLOW_TOKENS=1024 21 22# Set to "local" to skip the primary model and route all requests to the fallback. 23NIRI_ENV=default 24 25# Must match the user created in the Dockerfile and the volume mount in docker-compose.yml. 26NIRI_USER=niri 27NIRI_CONTAINER=niri 28# UID/GID the container process runs as — match your host user to avoid permission issues. 29AGENT_UID=1001 30AGENT_GID=1001 31 32PORT=3000 33# Used by the chat CLI to connect. Include the scheme, exclude the port. 34NIRI_HOST=http://localhost 35 36# ── Context & token limits ──────────────────────────────────────────────────── 37TOKEN_NUDGE_THRESHOLD=120000 38FALLBACK_TOKEN_NUDGE_THRESHOLD=50000 39CONTEXT_COMPACT_TARGET_TOKENS=65000 40CONTEXT_COMPACT_TRIGGER_TOKENS=90000 41CONTEXT_COMPACT_RECENT_MESSAGES=80 42CONTEXT_COMPACT_CHUNK_MESSAGES=32 43CONTEXT_COMPACT_SUMMARY_MAX_CHARS=16000 44 45# ── Tools ───────────────────────────────────────────────────────────────────── 46# Max bytes accepted by image_tool before rejecting the file. 47IMAGE_TOOL_MAX_BYTES=150000 48# Override the image directory exposed to the agent (default: /home/$NIRI_USER/images). 49IMAGE_ROOT= 50 51# ── Discord ─────────────────────────────────────────────────────────────────── 52DISCORD_BOT_TOKEN= 53# Optional — auto-detected from the gateway on first connect if left blank. 54DISCORD_BOT_USER_ID= 55# Enable the built-in gateway listener (messageCreate → ingest pipeline). 56DISCORD_GATEWAY_ENABLED=true 57# Log every ingest decision (verbose, for debugging). 58DISCORD_GATEWAY_TRACE=false 59# Also ingest raw MESSAGE_CREATE packets — recommended for DM reliability. 60DISCORD_GATEWAY_RAW_FALLBACK=true 61# Comma-separated channel IDs scanned by discord_scan when no channel_ids arg is passed. 62DISCORD_SCAN_CHANNEL_IDS= 63# Wake the agent immediately on any new Discord event (leave false to avoid spam). 64DISCORD_WAKE_ON_EVENT=false 65# Wake immediately on DMs even when DISCORD_WAKE_ON_EVENT is false. 66DISCORD_WAKE_ON_DM=true 67# How often to batch pending Discord events into a single harness message. 68DISCORD_BATCH_INTERVAL_MS=60000 69# Restrict batch digests to configured scan channels + DMs (recommended). 70DISCORD_BATCH_ONLY_CONFIGURED=true 71# Auto-demote pending inbox items to seen after N minutes (0 = disabled). 72DISCORD_PENDING_AUTO_SEEN_MINUTES=10 73# Include a channel scan snapshot in each batch payload. 74DISCORD_BATCH_SCAN=true 75# Max messages included in a single batch digest. 76DISCORD_BATCH_MAX_MESSAGES=40