syntropctl(1) — Unified Operator Manual

Global Options

Sends an org.varlink.service.GetInfo ping to all active sockets under /run/syntrop/.

DAEMON       SOCKET                                  STATE    LATENCY   INTERFACES
inferenced   /run/syntrop/io.syntrop.Inference1      ACTIVE    0.41ms   io.syntrop.Inference1, org.varlink.service
modeld       /run/syntrop/io.syntrop.Model1          ACTIVE    0.32ms   io.syntrop.Model1, org.varlink.service
contextd     /run/syntrop/io.syntrop.Context1        ACTIVE    0.28ms   io.syntrop.Context1, org.varlink.service
toold        /run/syntrop/io.syntrop.Tool1           ACTIVE    0.35ms   io.syntrop.Tool1, org.varlink.service
runtimed     /run/syntrop/io.syntrop.Runtime1        STANDBY   0.15ms   io.syntrop.Runtime1, org.varlink.service
routerd      /run/syntrop/io.syntrop.Router1         ACTIVE    0.38ms   io.syntrop.Router1, org.varlink.service

System Status: OPERATIONAL (All sockets listening; 0 MB idle RAM footprint)

Executes an end-to-end incident triage inquiry. Correlates unit failure signals with recent configuration drift, journal logs, and PSI pressure to pinpoint exact root causes.

● nginx.service - The NGINX HTTP and reverse proxy server
     Loaded: loaded (/usr/lib/systemd/system/nginx.service; enabled)
     Active: failed (Result: exit-code) since Wed 2026-09-24 04:52:10 UTC; 42s ago

INCIDENT CHRONOLOGY:
  ├─ 04:50:12: Package upgrade 'openssl-libs' (3.0.7 -> 3.2.0)
  ├─ 04:51:30: Config modified: /etc/nginx/nginx.conf (+ssl_protocols TLSv1.3)
  └─ 04:52:10: Process terminated with status 1 (nginx: [emerg] invalid directive)

ROOT CAUSE ANALYSIS:
  The directive 'ssl_protocols TLSv1.3' inside http {} block conflicts with
  upstream template syntax in /etc/nginx/conf.d/default.conf line 14.

SUGGESTED ACTION:
  # systemctl revert nginx.service
  Or run automated remediation:
  $ syntropctl run revert_config /etc/nginx/nginx.conf

PLANE ID           KIND       DEVICE NAME                 TOTAL VRAM   ALLOCATED    PRESSURE
drm-renderD128     DRM_GPU    NVIDIA GeForce RTX 4090     24.00 GiB     8.25 GiB    Normal (0.0% some)
cpu-amx-0          CPU_AMX    Intel Xeon Platinum 8480+   64.00 GiB     0.00 GiB    Normal (0.2% some)

Active Leases:
  lease-4921-bc  [Interactive]     payment.service   4.00 GiB  (plane: drm-renderD128)
  lease-7712-aa  [EmergencyTriage] sentry-enclave    4.25 GiB  (plane: drm-renderD128)

Pull models using aliases (qwen2.5:0.5b, qwen2.5:1.5b, llama3.2:1b, smollm2:360m, gemma4:e2b) or Hugging Face repositories (org/repo):

Resolving model: qwen2.5:0.5b
  Target: qwen2.5:0.5b
  Source: https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf
qwen2.5 [00:00:02] [########################################] 397.66 MiB/397.66 MiB (0s) Download complete
  SHA-256: 7f9a12bc81e3a4e98f01b3c4d5e6f708192a3b4c5d6e7f8091a2b3c4d5e6f708
Successfully pulled and registered qwen2.5:0.5b

Automatically size hardware envelopes, pull and pin draft & primary models via modelctl bootstrap, generate speculative router configurations, and reload services:

=== Syntrop Setup: Family [qwen] (Envelope: auto) ===
[1/3] Sizing hardware envelope and dispatching to modelctl bootstrap...
  CPU Cores: 12 | Host RAM: 14.8 GB avail (7.4 GB draft budget)
  Selected CPU Draft: qwen2.5:0.5b [Q8_0]
  [✓] modelctl bootstrap completed successfully.
[2/3] Generating speculative routerd.toml configuration...
  [✓] Speculative router configuration written to /etc/syntrop/routerd.toml
[3/3] Reloading routerd.service...
  [✓] routerd.service reloaded successfully.
Setup completed successfully for family: qwen

MODEL                        STATUS     SIZE         PARAMS     BACKEND      CONTEXT   
------------------------------------------------------------------------------------
qwen2.5:0.5b                 cached     397.66 MiB   0.5B       -            8192      
llama3.2:1b                  cached     1.24 GiB     1.2B       -            8192      
qwen2.5-coder-7b             loaded     4.48 GiB     7.0B       cpu-avx2     32768     

CAS Storage Quota: 6.12 GiB / 64.00 GiB (9.6% utilized) • 3 models (1 pinned)

TIME (UTC)         SUBSYSTEM   TARGET FILE / UNIT                       EVENT TYPE
04:51:30 (12m ago) contextd    /etc/nginx/nginx.conf                    CONFIG_MODIFIED
04:50:12 (13m ago) rpm-ostree  openssl-libs (3.0.7 -> 3.2.0)            PKG_UPGRADE
04:12:05 (51m ago) toold       /etc/systemd/system/redis.service.d/     DROPIN_RESTORED

routerd provides reverse proxy wire ingress on TCP port 32768 and Unix domain socket /run/syntrop/router.sock. Operators manage routing policies, test upstream providers, and verify model reachability using routerctl and standard HTTP utilities.

Verify Wire Ingress via curl

Verify that routerd is listening on port 32768 and responds to OpenAI-compatible model enumeration:

{
  "object": "list",
  "data": [
    {
      "id": "router:fast",
      "object": "model",
      "created": 1758784000,
      "owned_by": "syntropd-router"
    },
    {
      "id": "router:hard",
      "object": "model",
      "created": 1758784000,
      "owned_by": "syntropd-router"
    },
    {
      "id": "qwen2.5-coder-7b",
      "object": "model",
      "created": 1758784000,
      "owned_by": "runtimed-local"
    }
  ]
}

Inspect Router Status & PSI Pressure

● routerd.service - Intelligent Model Router and Wire Protocol Gateway
     Status: OPERATIONAL (Uptime: 1h 45m)
     Wire Ingress: 127.0.0.1:32768, /run/syntrop/router.sock
     Varlink IPC:  /run/syntrop/io.syntrop.Router1
     Memory RSS:   8.42 MiB (MemoryHigh: 24M, MemoryMax: 32M)
     Kernel PSI:   Memory 0.0% some (normal)
     Providers:    3 configured, 3 healthy

List and Test Configured Providers

PROVIDER ID       KIND     TIER    STATUS    WEIGHT   LATENCY    MODELS
runtimed-local    runtimed fast    HEALTHY   1.0      12.4ms     qwen2.5-coder-7b
lan-runtimed      runtimed fast    HEALTHY   0.8      48.2ms     gemma4:e2b, qwen2.5:7b
local-vllm        openai   hard    HEALTHY   0.9      62.1ms     qwen3:32b

[TEST] Probing provider 'runtimed-local' at unix:/run/syntrop/io.syntrop.Runtime1...
[OK] Provider response received in 14.2ms.
[OK] Health check passed: HTTP 200 OK (models: 1, state: ready)

Simulate Routing Decisions

ROUTING DECISION CANDIDATES:
  1. runtimed-local / qwen2.5-coder-7b  [SCORE: 96.4] Speed: 98.0  Cost: 100.0  Cap: 90.0  (Selected)
  2. lan-runtimed / gemma4:e2b          [SCORE: 81.2] Speed: 72.0  Cost: 100.0  Cap: 75.0
  3. local-vllm / qwen3:32b             [SCORE: 64.8] Speed: 42.0  Cost: 30.0   Cap: 99.0

1. Constrained JSON Grammar Decoding

Force model generation to adhere strictly to JSON schemas or Varlink payload syntax using SIMD-accelerated logit masking:

varlinkctl call unix:/run/syntrop/io.syntrop.Runtime1 io.syntrop.Runtime1.Generate \
  '{"model":"qwen2.5:0.5b","prompt":"Emit error report:","grammar_type":"json","grammar":"{}","max_tokens":128}'

2. Heterogeneous Speculative Verification

Gang-schedule a CPU draft model (e.g. qwen2.5:0.5b) with a GPU target model (e.g. qwen2.5:7b) for 2x–3x generation acceleration:

varlinkctl call unix:/run/syntrop/io.syntrop.Runtime1 io.syntrop.Runtime1.Generate \
  '{"model":"qwen2.5:7b","speculative_draft_model":"qwen2.5:0.5b","prompt":"Analyze cgroup freeze:","max_tokens":256}'

3. Reasoning Budget & Wire Think Filter

Specify reasoning token limits via OpenAI-compatible HTTP ingress, streaming reasoning chains separately from final content:

curl -N -s http://127.0.0.1:32768/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "router:fast",
    "reasoning_budget": 256,
    "stream": true,
    "messages": [
      {"role": "user", "content": "Triage degraded disk I/O on /var/log"}
    ]
  }'

4. Infinite Context StreamingLLM

Streaming journal monitoring runs 24/7 with pinned attention sinks and sliding FIFO windows in SinkWindowCache, preserving bounded RAM.

5. Multimedia Audio & Visual Generation

Generate real-time audio streamed to PipeWire or synthesize 1-step SD-Turbo visuals via isolated Varlink compute leases:

varlinkctl call unix:/run/syntrop/io.syntrop.Runtime1 io.syntrop.Runtime1.StreamAudioOut \
  '{"text":"System alert: cgroup CPU throttling exceeded threshold","voice":"af_heart","rate":24000}'

6. Dynamic Memory Telemetry & Leases

Query DRM device watermark levels and inspect dynamic memory leases:

varlinkctl call unix:/run/syntrop/io.syntrop.Inference1 io.syntrop.Inference1.GetDrmWatermark \
  '{"device":"renderD128"}'

7. Accelerated Hardware Pipeline (Marlin GEMV, Ada Lovelace FP8 & Paged KV)

runtimed incorporates an ultra-fast hardware acceleration pipeline featuring:

• In-VRAM 4-Bit Marlin GEMV: Repacks weights into 128-bit Marlin INT4 tiles executed via custom asynchronous PTX CUDA kernels (marlin_gemv_kernel), delivering sub-millisecond per-token generation latencies.
• Ada Lovelace FP8 Tensor Core Pipeline: Hardware-native F8E4M3 matrix multiplication with dynamic per-channel scaling factors targeting NVIDIA Ada Lovelace / Hopper Tensor Cores.
• Quantized Paged KV-Cache: 16-token paging blocks with per-head FP8 quantization, shrinking KV-cache memory pressure by 50% without quality degradation, backed by kernel PSI asynchronous host RAM spill/unspill management.
• Dual-GPU Speculative Offload: Seamless gang-scheduling across multiple GPUs (e.g. primary target on cuda:0 and draft engine on cuda:1).

Add a standard systemd drop-in file to your service (e.g. /etc/systemd/system/app.service.d/triage.conf):

[Unit]
# OnFailure automatically invokes the syntrop-triage template unit passing the failed unit name (%n)
OnFailure=syntrop-triage@%n.service

When app.service enters the failed state, systemd PID 1 starts syntrop-triage@app.service.service:

[Unit]
Description=syntropd Autonomous OS Self-Healing & Remediation for %I
Documentation=https://syntropd.github.io/manual.html
After=syntrop-sockets.target sentry.service toold.service contextd.service

[Service]
Type=oneshot
User=syntrop-admin
Group=syntrop
ExecStart=/usr/bin/syn admin remediate %I
StandardOutput=journal
StandardError=journal
SyslogIdentifier=syntrop-admin

# Systemd Sandboxing
ProtectSystem=strict
ProtectHome=true
PrivateTmp=true
NoNewPrivileges=true