A project for comparing model choices. Daily scoring is not yet armed.
Tugboat is an internal project described as a model-selection map. Status recorded in the October 6, 2026 audit: its runner and scorer are reported built and runnable, the daily schedule is not armed, and live production routing is not active. The page does not show every data path.
Tugboat is described as a way to compare model choices by task and budget. The page lists public benchmark and pricing inputs, but it does not show daily results or a live routing action.
The name comes from the job. A tugboat doesn't build the ship or plan the route. It guides heavy things into the right position efficiently and gets out of the way. The design puts this map beside Hermes, an internal conductor-agent project, and Pilot, an explainer of checks for AI-assisted code work.
Concept diagram: the page lists public benchmark and price inputs, an internal usage log, and a model map for Council review. It does not show a live result, armed schedule, or complete data-flow record.
How It Works
Public Data In, Routing Map Out
The page lists public benchmark and price sources plus an internal usage log as intended inputs. It describes local scoring but does not show the data path for every model, account, or run.
Input
Data Sources Described
This page lists public benchmark results, model pricing, and an internal usage ledger as data sources. It does not verify where data goes in every run.
Scoring
Task-Class Scoring
The page describes scoring model choices by task class and budget, then presenting a ranked list to the Council. It does not publish the scoring formula, a sample ranking, or a run record.
Cadence
Daily Schedule Not Armed
The page reports that a daily schedule is planned but not armed. It does not show a daily run, a manual review cadence, or a dated model map.
Governance
The Council Still Governs
Lighthouse Council is Lighthouse Agentic's internal group for reviewing model choices and setting capability rules. This page describes the map as an input to that review, not an automatic switch.
The Thinking Behind It
What This Page Describes
The page describes intended inputs and a local-scoring design. No sample map, run log, or complete data-flow record is published here.
The page names public sources. It lists benchmark scores and pricing as inputs. It does not show whether other data is sent in every setup.
Local scoring is described. The page says rankings are calculated locally and written to a file. It does not show every place data goes during a run.
Source gaps matter. The design says a missing source should be flagged before a routing choice. This page does not include code or a run record that verifies that behavior.
Live routing is a later step. The page says production routing waits for a second source to confirm the public benchmark picture. It does not show a live route or the source record.
Where It Stands
Runnable Baseline; Routing Not Live
Status recorded in the October 6, 2026 audit. These are page-reported states; the page does not include a run record.
Page-reported status
Architecture and Scoring
The page reports a runner and scorer that are built and runnable. It says the daily schedule is not armed. The page does not publish a sample run or show where data goes in every setup.
Next gate
Second Source Is a Gate
The page says production routing is gated on a second source. It does not show which source is active or how its quality is checked.
Runner + scorer: reported built and runnable
Daily scoring: scheduled (not yet armed)
Live production routing: next gate
Tugboat is an internal project, not an external product offer. This page describes its intended role for the Lighthouse Agentic Council. For a conversation about our work, Start a Conversation.
The Right Tool for the Job
Compare choices. Keep a person in the decision.
Tugboat is described as an internal model-selection map. Daily scoring is not armed, and live production routing is not active.