Everything you need to build serious AI
No fragmented vendors. No duct tape between tools. One company shipping the full stack — and adding more.
Govern the agents you're shipping to production
Four products on one data layer. Every tool call traced, enforced, evaluated, and mapped to compliance evidence — automatically, not on audit day.
Observe
See every AI action
- Live activity feed across every agent
- Inventory: applications, agents, MCP servers, datasets
- Full causal trace for every tool call
Guardrails
Block what shouldn't happen
- Visual rule builder with plain-English preview
- Block, hold, redact, flag — at call time
- Inline Create policy from any uncovered call
Evaluation
Score what did happen
- Taxonomy-driven QA, staged AI auditor
- Human, assisted, or fully automated review
- Training data export: SFT / DPO / ranking
Governance
Prove what happened
- EU AI Act · NIST AI RMF · ISO 42001 · SOC 2
- Evidence auto-mapped from runtime decisions
- Signed audit bundles on demand
Auto-mapped evidence, not another PDF audit
Every tool call. Every decision. Fully visible.
Install once. Call the gate before a tool runs. Bookbag decides allow, block, hold, or redact — and every decision feeds the activity feed, the compliance bundle, and the eval dataset.
# pip install bookbag
import os
from bookbag import BookbagClient
client = BookbagClient(api_key=os.environ["BOOKBAG_API_KEY"])
run = client.agent.start_run(agent_id="support-agent")
# Every tool call goes through Bookbag — no policy? it's logged.
# Policy fires? it's blocked, held for review, or redacted.
result = client.agent.tool_call(
run_uid=run["run_uid"],
tool="stripe.issue_refund",
arguments={"amount": 500.00, "order_id": "FF-4210"},
)
# → {'decision': 'hold', 'reason': 'refund > $100 requires approval',
# 'audit_id': 'aud_...', 'run_uid': '...'}query_databaseallowcompletedsend_emailblockcompletedissue_refundholdpendingissue_refundallowcompletedupdate_crmallowcompletedfetch_piiblockcompletedOpenBook — an open-source web AI agent. Yours to keep.
OpenBook is an AI agent that runs in the browser. No local install. No vendor lock-in. Self-host it on your own infrastructure or use our hosted version. Fork it, extend it, ship it.
- Browser-native — open a tab, run an agent
- Self-host with Docker in under five minutes
- MIT licensed — fork it, it's yours
- Built and maintained by the Bookbag team
3 risks identified:
1. Auto-renewal clause — 90-day cancellation window
2. Liability cap set at 1× annual fees, not damages
3. Governing law: Delaware, disputes require arbitration
We keep building. That's what we do.
Governance is the flagship. OpenBook just shipped. We're not done. If you're working on something that needs AI infrastructure we haven't built yet — reach out.
Talk to us about what you're buildingStart with the OSS SDK. Scale to enterprise.
Four tiers for the Bookbag governance platform — from the free MIT-licensed SDK all the way to on-prem with an SLA.
Frequently asked questions
Can't find what you're looking for? Reach out.
Frequently Asked Questions
We're building the AI company. Come build with us.
Join the teams shipping safer AI with real-time evaluation, audit trails, and continuous improvement.