recall vs Bito

The overlap is the diagnosis, not the fix. Both products target context waste in coding agents, but Recall solves it by remembering your own session, and Bito solves it by pre-loading your codebase. Recall captures each session through Stop/SessionEnd hooks, appends only new turns to a plaintext history.md (prompts, replies, files touched, commands run), and condenses it with a local TF-IDF + TextRank pipeline (scripts/summarizer.py) into a ~1–2K-token context.md that the SessionStart hook surfaces with a resume offer. That summarization is extractive and deterministic — facts come from the transcript and git diff --stat, never an LLM call, so it spends zero model tokens. Support is limited to Claude Code and Opencode, with auto_save_context: "on_end" and a manual /recall:save checkpoint command.

Bito Governor changes the request itself rather than the session log. Its Code Context Engine builds a knowledge graph of the codebase and serves a distilled brief of relevant files, symbols, and dependencies with every request, plus complexity scoring against services touched, dependency depth, and blast radius to decide routing. It spans Claude Code, Cursor, Codex, GitHub Copilot, and Pi, streams and preserves tool calls through the routing hop, covers Anthropic, OpenAI, Gemini, Grok, and open-weight models, and adds budgets per team or key with token/spend analytics. Bito publishes model-comparison research (Sonnet 5.5 vs Sonnet 5, Opus 5.5 vs Opus 5) to inform routing. The trade-off is scope and setup: Bito claims steps per task falling 47→23 and a 10-task session dropping 25.3→8.5 minutes, but indexing a large codebase is not a same-day job. Recall installs and logs immediately because it never has to understand your whole repo.

Recall is free — MIT-licensed, with no API key, no external model, and nothing sent anywhere. The cost model is opportunity, not dollars: you stop paying usage credits to re-explain yesterday's work, and summarization runs locally so refreshing memory spends no model tokens. For a solo dev on a Claude Code or Opencode subscription, that's a pure win.

Bito Governor is freemium and priced as a spend-reduction tool, not a subscription sidecar. It advertises AI model discounts of 15–25% off list price, and you can either bring your own provider keys or run it alongside an existing gateway — so it's designed to sit in a bill that's already large enough to optimize. The real gate is the not-for list: if token spend is too low to justify scoping, a plain gateway will look cheaper per token. Bito's own 2026 research pushes exactly this framing: per-token pricing no longer predicts total agent spend, so measure against your live traffic (on/off savings measurement is built in) before committing. Teams get budgets per team or per key with token and spend analytics in one admin view; routing policies and quality floors are Enterprise. Practically: Recall's price is $0 and it never touches procurement. Bito needs a big enough multi-repo bill to clear its setup cost — the 78%-of-bill-goes-to-context-retrieval argument is the pitch you'd take to whoever signs off.