Luma Operator Control Plane

Proof of life. Truthful readiness. No guesswork.

This page is served by the live gateway process on this host. It reflects the current profile, current service posture, and the only supported connect paths in this wave: Telegram and webhook ingress.

Needs attention Â· 1 required issue(s)

Operator URLhttp://127.0.0.1:8787/

Checked2026-06-27T21:00:59.329691922+00:00

Runtime

Service posture

Installed: yes

Scope: system

Status command: systemctl status luma.service

Logs: journalctl -u luma.service -f

Manual start: luma daemon

Profile

Filesystem anchors

Profile root: /home/claw/.luma

Config: /home/claw/.luma/config.toml

Secrets: /home/claw/.luma/secrets.toml

Workspace: /home/claw/.luma/workspace

Ready marker: yes

Webhook

Gateway ingress

Bind: 127.0.0.1:8787

Exposure: public_tunnel

Auth: bearer

Generic: http://127.0.0.1:8787/webhook

Nextcloud Talk: http://127.0.0.1:8787/nextcloud-talk

Tunnel: tailscale_funnel (listen Some("443"), hostname None, domain None)

Provider

Default LLM route

Route: default

Provider: Local OpenAI-compatible

Model: gpt-5.4

Base URL: https://api.ikuncode.cc/v1

Support: Production-supported providers in this wave: OpenAI, Anthropic, and local OpenAI-compatible endpoints.

Health: error: openai backend error: OpenAI responses provider rejected request (status=401 Unauthorized, request-mode=sse, content-type=application/json; charset=utf-8, content-encoding=<unknown>, transfer-encoding=<unknown>): Invalid token (request id: 202606272059163760519128268d9d6afalkvim)

verify the default LLM provider credentials, model name, base URL, and outbound network access

TDLib route

Configured: yes

Mode: bot

Status: ready

Auth state: authorizationStateReady

Session dir: /home/claw/.luma/state/tdlib

Send a Telegram message from an allowed handle to confirm the live route.

Companion

Memory runtime

Mode: local_managed_stack

Status: healthy or ready for cutover

Use `luma companion prepare-graphiti --start` and rerun doctor until the managed stack is healthy.

Telemetry

Task and trace visibility

Active root tasks: 0

Recorded LLM calls: 2654

Tokens in/out/total: 5295233/728201/6023434

Last trace: c585e9b0-5a73-4270-8fda-f5a5d03afdd3

Last request id: n/a

Degraded memory visible: no

llm provider request failed: openai backend error: OpenAI responses provider rejected request (s [failed]
updated 2026-06-27T19:46:01.029980083+00:00 · trace c585e9b0-5a73-4270-8fda-f5a5d03afdd3 · request n/a

llm provider request failed: openai backend error: OpenAI responses provider rejected request (s [failed]
updated 2026-06-27T18:56:58.386830382+00:00 · trace bb3b82a0-142a-4a45-bfee-02b664115975 · request n/a

llm provider request failed: openai backend error: OpenAI responses provider rejected request (s [failed]
updated 2026-06-27T18:29:58.067806161+00:00 · trace 893bb8f4-1fc9-4b4c-bf72-01997f21d074 · request n/a

llm provider request failed: openai backend error: OpenAI responses provider rejected request (s [failed]
updated 2026-06-27T18:11:58.236767512+00:00 · trace 68d80c49-cc3f-4c76-b465-4f8738482d76 · request n/a

Readiness

Doctor snapshot

Required failures: 1

Overall: blocked

Refresh http://127.0.0.1:8787/operator/status for JSON or rerun luma doctor for the full CLI report.

Connect checklist

Only the supported channels

ready

mode=bot session_dir=/home/claw/.luma/state/tdlib

Send a Telegram message from an allowed handle to confirm the live route.

Webhook

waiting for first request

Listener is live, but no webhook traffic has reached this gateway yet. POST JSON to http://127.0.0.1:8787/webhook (Nextcloud Talk: http://127.0.0.1:8787/nextcloud-talk); auth=bearer. Listener=listening timeout/failure counts=0/0.

Send a webhook request to the local gateway, then refresh this page.

Checks

Operator-facing readiness detail

Check	Status	Result	Next step
workspace	PASS	ok	â€”
ready_marker	PASS	ready marker matches this profile	â€”
owner_binding	PASS	owner binding is healthy for Telegram user 56483202	â€”
llm	FAIL	error: openai backend error: OpenAI responses provider rejected request (status=401 Unauthorized, request-mode=sse, content-type=application/json; charset=utf-8, content-encoding=<unknown>, transfer-encoding=<unknown>): Invalid token (request id: 202606272059163760519128268d9d6afalkvim)	verify the default LLM provider credentials, model name, base URL, and outbound network access
llm_memory_route	PASS	background memory/reflection use explicit llm route [memory_writer_default -> memory_writer_json_fallback]	â€”
llm_responses_transport	WARN	custom OpenAI-compatible gateway routes explicitly use SSE-first transport: default (https://api.ikuncode.cc/v1), final_synthesis_default (https://api.ikuncode.cc/v1), memory_writer_default (https://api.ikuncode.cc/v1)	set llm.providers.<route>.responses_transport = "json" or "auto" unless the gateway is known to support stable Responses SSE
llm_tool_budget	PASS	configured single-turn tool budget = 50	â€”
llm_vision	FAIL	error: openai backend error: OpenAI responses provider rejected request (status=401 Unauthorized, request-mode=sse, content-type=application/json; charset=utf-8, content-encoding=<unknown>, transfer-encoding=<unknown>): Invalid token (request id: 202606272059172394319238268d9d6WKNBwVah)	verify that the default LLM route supports image inputs on the selected endpoint/mode, then rerun `luma doctor --repair` to auto-enable the vision flags
embeddings	PASS	ok via ollama:http://127.0.0.1:11434:dengcao/Qwen3-Embedding-4B:q4_k_m:2560/dengcao/Qwen3-Embedding-4B:q4_k_m (dims=2560)	â€”
sqlite	PASS	ok	â€”
llm_attempts	WARN	recent_attempts=100 failed=100 failed_missing_decision=0 gateway_502=0 one_attempt_failures=98 failover_available=true configured_providers=6 sse_read_or_parse_failures=0 sse_2xx_semantic_parse=0 stream_read_error=0 api_families=responses:100 endpoints=/v1/responses:100 transports=json:2,sse:98; last_failure api_family=responses endpoint=/v1/responses transport=sse http_status=401 stage=provider_rejected decision=non_retryable_provider_rejection code=none	inspect get_llm_trace attempts; provider failures should normally show retry decisions unless the error is non-retryable
production_compaction_backlog	WARN	routes_over_threshold=1 unsummarized=1 threshold=128 top=[telegram:56483202:none:none raw=235 summaries=0]	run/verify context compaction for noisy routes before retrying high-budget autonomy tasks
production_retry_amplification	WARN	repeated_domains=1 repeated_failures=48 browser_stale_failures=0 browser_stale_durable_refs=0 top=[provider_failure:openai\|gpt-5.4\|responses\|/v1/responses\|sse\|provider_rejected\|none\|401 count=48 tasks=42]	stop blind retries for repeated signatures; inspect evidence and surface typed failure or final synthesis
companion_memory_writes	WARN	recent_outcomes=50 succeeded=19 failed=31 zero_write_failures=31 unclassified_failures=0 incomplete=0 stale_unclassified=0 retryable=17 transport_failures=4 companion_episodes=70 companion_pinned_states=57 companion_memory_patches=793 conversation_topics=0; last_failure source_task_id=576bbfd8-4ab1-43a0-9c3f-7bbfed51c9e3 class=provider retryable=false	recent companion memory writes failed; inspect the classified failure reason before treating recall as healthy
identity	PASS	ok (aieos)	â€”
companion_persona	PASS	ok (persona_slug=luma persona_id=77b5a08d-b6ee-5e2c-a661-feb6327b274f)	â€”
memory_runtime	PASS	ok	â€”
companion_semantic_recall	PASS	pgvector-backed semantic recall is available	â€”
companion_memory_truth	PASS	graphiti-backed truth is healthy	â€”
skills	PASS	0 active, 0 promptable, 0 always-on	â€”
plugin_runtime	SKIP	skipped	â€”
browser	PASS	local browser automation ready (browser=/usr/bin/chromium chromedriver=/usr/bin/chromedriver)	â€”
gateway	PASS	ok (public_tunnel)	â€”
tdlib	PASS	ok	â€”
companion_cutover	PASS	ok	â€”
companion_docker	PASS	ok via unix socket /var/run/docker.sock	â€”
companion_graphiti	PASS	ok	â€”
companion_graphiti_search	PASS	ok	â€”
companion_neo4j	PASS	ok via HTTP 200 OK	â€”
companion_postgres	PASS	ok	â€”