foreman vs Bito

Foreman enforces process. It spawns your local claude CLI in headless stream-json mode and drives a fixed pipeline — plan, ADR/PRD, issues, TDD build, e2e — with a human review gate at every design phase and a PreToolUse deny hook that stops workers writing their own verification. Approvals are hash-sealed and auto-revert if a design doc changes; per-run turn, cost and time budgets apply to every worker, with a daily ceiling that hard-stops. Workers get worktree isolation bounded by a declared touches set, and state lives as human-readable files committed in the target repo. A demo mode runs the full pipeline on canned stream-json at zero token cost, and an evals flywheel (retro clusters failures into gated patches; bench replays the eval set reporting success-rate, cost and turn deltas) closes the loop. Bito solves a different problem: a drop-in endpoint that scores each request's structural complexity — services touched, dependency depth, blast radius — and routes it to a right-sized model, while a Code Context Engine hands the model a distilled brief of relevant files, symbols and dependencies. It preserves streaming and tool calls, works with your own keys or alongside an existing gateway, and covers Anthropic, OpenAI, Gemini, Grok and open-weight models. Where Foreman decides how work proceeds, Bito decides which model does it and what context it sees.

Foreman is free — MIT-licensed, with no licence fee in the data. Your spend is whatever Claude Code consumes, but Foreman is unusual in bounding it: per-run turn, cost and time budgets plus a daily hard ceiling, and a demo mode that exercises the full pipeline at zero token cost. Bito is freemium: the router is the product and you pay for the decision layer. It advertises AI model discounts of 15-25% off list price, and when you bring your own provider keys it claims 40-70% cuts on coding-agent spend, with continuous on/off savings measurement against your own live traffic — so the justification is a measured delta, not a promise. Enterprise adds custom routing policies and quality floors per key, plus per-team or per-key budgets and token/spend analytics in one admin view. The pricing logics don't overlap: Foreman charges nothing and limits consumption; Bito charges for reducing consumption on agents you already run. A shop pinning one model everywhere and shopping purely on price per token will find a plain gateway cheaper than Bito.