Bullswarm can inventory the models exposed by installed agent CLIs and combine connector-declared, dated pricing/benchmark metadata with live quota surplus. Effort tiers are high, medium, and low. A pool is one installed agent CLI, or one account of that CLI.

bullswarm setup                            # TTY: interactive control center
bullswarm strategy                         # explicit routing-focused alias
bullswarm strategy inventory --json        # agent-readable detection + policy + routes
bullswarm strategy routes --json           # compact effective choices
bullswarm strategy set-provider codex off --yes
bullswarm strategy set-model opencode2 relay/gpt-5.6-luna \
  --tiers high,medium,low --yes
bullswarm strategy configure --file strategy.json --yes  # atomic agent-created policy
bullswarm strategy reset-tier low --yes     # restore one tier to automatic
bullswarm strategy set-reasoning --tier high --level xhigh --yes
bullswarm strategy set-reasoning --tier high --level high --pool codex --yes
bullswarm strategy reset-reasoning --tier high --yes  # back to connector defaults
bullswarm strategy refresh
bullswarm strategy show --json
bullswarm strategy apply --yes --refresh-hours 24
bullswarm strategy auto status
bullswarm strategy set-subscription command-code \
  --plan GOAT --monthly-usd 10 --included-usd 70 --quota-window monthly
bullswarm strategy set-subscription opencode2 \
  --resets-at 2026-09-17T01:46:01Z   # usage is reported, the refill date is not
bullswarm strategy assign high --pool claude-code --model claude-opus-4-6
bullswarm strategy exclude-model claude-fable-5
bullswarm run --effort high --lane analyze --task-file /tmp/task.md --json

Rungs

A rung is one pool’s model plus its reasoning level for one effort tier — the two halves you actually choose together. bullswarm strategy rungs reads them as one table and bullswarm strategy set-rung writes both halves in one atomic save:

bullswarm strategy rungs                      # every enabled pool x configured tier
bullswarm strategy rungs --json --pool codex  # machine-readable, one pool
bullswarm strategy set-rung codex high --model gpt-5.6-sol --reasoning xhigh

Each row carries the effective model and where it came from, the effective reasoning level and which layer chose it, the dated benchmark evidence for that model at that reasoning level (blended, $/task, tok/task), and what this machine recorded for that pool and tier (dispatch count, median wall minutes, ok share). Evidence and record are never estimated: a model the datapack does not cover prints no evidence, and a tier with no matching attempt prints no dispatches. strategy inventory --json carries the identical rows under rungs.

Reading is free of side effects — no state write, no model discovery, no download. set-rung never spawns discovery either: a model absent from the pool’s cached discovery exits 2 and lists the models it does know, unless you pass --force. A reasoning level the connector cannot express is clamped down to the strongest level it accepts and the clamp is printed. A rung is singular per pool and tier, so the tier moves off whichever model held it while that model keeps its other tiers. Nothing about state.json changed shape: rungs are a view over strategy.modelTiers and strategy.reasoning.

The Relay-backed OpenCode pools (opencode2, opencode2:relay-2, opencode2:relay-3) express reasoning as opencode’s --variant <level>, at the same five levels as command-code. opencode only forwards a variant its own config declares for that model, so bullswarm injects them: each pool is spawned with OPENCODE_CONFIG_CONTENT declaring low/medium/high/xhigh/max as reasoningEffort variants of <providerId>/gpt-5.6-luna, merged over your ~/.config/opencode/opencode.json (your API keys stay in force). Without that injection opencode accepts --variant and silently drops it. Set a rung per pool, using that pool’s own provider prefix:

bullswarm strategy set-rung opencode2:relay-2 medium \
  --model relay-2/gpt-5.6-luna --reasoning max

If you set OPENCODE_CONFIG_CONTENT yourself in ~/.bullswarm/connectors/opencode2.json, bullswarm leaves it alone and injects nothing — you own the variants from then on.

The benchmark evidence comes from Epoch AI’s benchmarking hub, used under CC BY 4.0: Epoch AI, ‘AI Benchmarking Hub’. Published online at epoch.ai. Retrieved from https://epoch.ai/benchmarks. blended is the mean of the cursorbench, deepswe, arc-agi-2, and critpt scores recorded for that exact model and reasoning level; cost and tokens per task come from cursorbench. See data/README.md for the schema and the refresh job.

Setup first asks whether to analyze live usage and recommend routes or open the current configuration for manual editing. Analysis shows a spinner plus per-provider usage progress, then presents the proposed defaults before making any routing change. Press Y to apply them or N to retain the current policy. The analysis selects at most one default model for each provider and effort tier. It uses OpenRouter’s agentic, coding, and intelligence indices as quality signals and API-equivalent pricing as the budget signal. A repository-owned benchmark refresh job refreshes two public assets on the rolling benchmark-data-latest GitHub Release: openrouter-benchmarks.json from the authenticated OpenRouter APIs, and epoch-benchmarks.json from Epoch AI’s CC BY 4.0 benchmark export, which is what strategy rungs reads for per-model-per-reasoning-level evidence. Installed CLIs download only those public files and never need or receive an OpenRouter key. The sources are OpenRouter’s benchmarks API and models API. The CLI caches each datapack under ~/.bullswarm/cache/. OpenRouter is cache-or-network only: a fresh cache is used as-is, otherwise the rolling release is fetched, and a cache miss with no network yields an empty catalog plus connector metadata — there is no bundled data/openrouter-benchmarks.json. Epoch keeps data/epoch-benchmarks.json as a bundled last-resort, so a missing network never blocks setup when that file exists.

The TUI lists every detected provider/account separately so its toggle matches its own quota meter. Enter drills into that provider’s detected models. In the model matrix, Up/Down selects a model, Left/Right moves a visibly highlighted cell across High, Medium, and Low, and Enter toggles that cell. Type to filter model names; assigned models sort above unassigned or disabled models. Select Finish setup and press Enter, or press F directly, to leave the control center. The effective-route panel is recomputed from the same policy and live surplus used by real dispatch. Provider and model edits affect new direct runs and workflow dispatches.

An external AI agent should first read strategy inventory --json, then use the validated set-provider / set-model commands or write one JSON document for strategy configure --file. Unknown pools and models are rejected before state is saved. Existing automatic choices are preserved when a human begins curating a tier; a model-level off never empties unrelated tiers.

{
  "providers": { "codex": false, "opencode2": true },
  "models": {
    "opencode2": {
      "relay/gpt-5.6-sol": ["high"],
      "relay/gpt-5.6-luna": ["medium", "low"]
    }
  }
}

Interactive setup asks whether to enable strategy autopilot; non-interactive setup requires the explicit setup --yes --strategy flag. Recommendations are context-filtered before ranking: high requires analysis plus workflow-planning, medium requires build/edit capabilities, and low targets bounded chores. An approved policy refreshes stale discovery before later runs and re-applies the best eligible models on its configured interval. Disable it with strategy auto off --yes. Discovery commands, model argument syntax, pricing, and benchmark declarations remain connector-owned. Unknown license value, prices, and benchmarks stay null rather than being guessed.

How an assignment interacts with pace, 5-hour headroom, and load at dispatch time — and the routing reason and candidate rows that explain a pick — is covered in Routing, not repeated here.

Model exclusions are hard routing policy. An excluded model is removed from recommendations and assignments, and Bullswarm pins a same-tier allowed model through the connector-owned model flag whenever the provider default could be excluded. A pool that cannot guarantee the exclusion is ineligible for that dispatch. Reverse the policy with bullswarm strategy include-model <model>.

Every run and workflow attempt reports its selected agent/model and estimated usage. When a delegate does not expose counters, Bullswarm labels its UTF-8 byte/4 token estimate. The breakdown separates standard read, cache read, cache write, and output; API-equivalent cost and normalized subscription quota remain unknown unless the connector and user-provided subscription data can support them. workflow tui --json <id> exposes the aggregate and the full phase/step/attempt tree.