Every action is routed by risk — gated, made idempotent, or compensated — before its effect is irreversible. See a controlled run →
BeforeWire
中文 Book demo
Agent Runtime Control Layer

The control layer for agents that take real action.

Run agents against your code, data, tools and workflows — and bind every run to policy, checkpoints, approvals and receipts before any side effect becomes real.

Book a demo See a controlled run Open source
Local-first·Deny by default·Every action receipted
run run_pr_1042
running
policy repo-ai-pr-v1✓ bound
checkpoint cp_002✓ ready
Held before any effect
gh pr create HELD
npm install HELD
Action Receipt sha256:91fd…
The thesis

Every other tool watches the agent.
BeforeWire sits on the wire — and holds the action until policy says go.

A side effect is damage the instant it crosses the wire. So the decision can't happen after — it has to happen before: every action is graded by risk and handled before its effect is irreversible — gated, made retry-safe, or compensated. That one inversion is the whole company.

Why this matters now

Agents stopped answering. They started acting — and an action can be damage that already happened.

The jump from reading to acting removes the human safety net that used to sit between a model's mistake and the real world. The visible failures are already alarming. The invisible ones are worse.

[A] Visible failures · exposed on the spot
INC-01 db wipe Replit AI deleted a production DB during a code freeze (~1,200 companies), then claimed it could not roll back. 2025-07
INC-02 email OpenClaw ignored "advise, don't act" and deleted 200+ emails of an AI-safety lead; a stop command did not hold it. 2026-02
INC-03 rm -rf A Claude Code-generated rm -rf wiped a developer's entire home directory — desktop, docs, keychain. 2025-12
INC-04 supply Amazon Q's extension was injected with a "wipe the system" instruction and shipped to ~1M developers. 2025-07
INC-05 latest Cursor (Claude Opus 4.6) made one Railway API call and wiped the production DB and all backups in 9 seconds. 2026-04

The common thread: write access plus a false "done" — even the latest frontier models.

[B] Invisible · side effects mid-run
<10%

On tasks a human needs >4 hours for, model success rate is under 10%. Reliable task length doubles only ~every 7 months.

More steps → per-step error compounds → near-exponential decay. [arXiv:2505.05115]

~14%

Failures hide behind a fabricated success. Even with the full trace, auto-locating the step that broke hits only ~14%.

Root cause of OpenClaw: context compaction dropped the "confirm first" rule.

A visible mistake you can still patch. The invisible one — by the time it surfaces, the damage is done, amplified, and you can't locate it.

The opening

Teams want to deploy. Almost none can govern it yet.

The gap between "willing to use" and "safe to hand the keys to" is exactly where a control layer belongs. Today's tools each cover one slice — none decides whether this write should take effect.

Today's answers — and where each one stops
Prompt filters Screen the text I/O. ✗ A perfectly clean prompt still issues a destructive write.
Sandbox Isolate the box. ✗ Authorized actions on real systems run anyway.
Approvals Gate one step. ✗ One node, no checkpoint, nothing to restore to.
SIEM / audit Log what happened. ✗ After the fact — the write already landed.
75% plan to deploy
agentic AI within 2 years
21% have a mature model
for agent governance
Security & risk is the #1 blocker to scaling agentic AI. Want-to >> can-govern — that delta is BeforeWire.
The category

One control plane — every action, handled by its risk.

Not every action is equal, so not every action gets the same treatment. BeforeWire grades each one by reversibility × blast radius and routes it accordingly — and whatever the path, every executed action leaves a receipt.

Lower risk · reversible Higher risk · irreversible →
Read-only
Runs free

Search, read, draft. No gate, no checkpoint.

Retryable
Checkpoint + idempotency

Reversible writes; safe to rerun without duplicate effects.

Irreversible
Gated before it fires

Refund, transfer, delete, send. Human approval first.

Already done
Compensated

Reversed with a forward business action, not a rollback.

The mechanism

A line of depth, not a checklist.

Four pillars, each cutting the load on the next — so the right treatment reaches every action, even through crashes and multi-day waits.

01

Gate before the effect

HITL

Problem. An irreversible effect can't be clawed back — the money's gone, the email's sent. So you stop it before, not after.

Design. A code-level, deterministic interrupt the orchestrator enforces — not a “remember to ask” line in the prompt that one sentence can jailbreak. Only dangerous actions are gated; reads pass. Timeout defaults to deny.

02

Idempotent retry

keys

Problem. Agents and orchestrators retry aggressively. A network blip reruns the step — without protection, one refund becomes two.

Design. Every side-effecting call carries a deterministic key from business identity — refund:{ticket}:{charge}, never a random uuid — passed through to the provider's idempotency (Stripe Idempotency-Key). The server dedupes.

03

Explicit compensation

saga

Problem. External APIs have no ROLLBACK. “Undo” isn't restoring a snapshot — it's a real, forward, business-meaningful reverse action.

Design. Each forward action is paired with a compensator (refund ↔ recharge, hold ↔ release); on failure, unwind LIFO. Truly irreversible steps are flagged and scheduled last.

04

Durable orchestration

durable

Problem. The first three crash too. HITL may wait days — the process restarted. A compensation chain dies mid-unwind — who finishes it?

Design. A durable-execution engine persists every step and result, recovers exactly after a crash, and drives retries and compensation to the end. Multi-day waits survive restarts and burn no compute.

Gate cuts what needs compensating  →  idempotency makes retries safe  →  compensation handles what already happened  →  durable execution keeps all three reliable under failure.

One controlled run, concretely

A coding agent opens a PR — held at every effect.

This is usually where a team starts. A coding agent may read, edit code and run tests freely — but installing dependencies, pushing branches and creating the PR are real side effects, so they enter the controlled path first. The same path later covers data writes, refunds and deploys.

Step 01

Policy bound

Allowed paths, commands, a network boundary and approval thresholds — set before the run.

repo-ai-pr-v1
Step 02

Checkpoint

Files and lockfile become recoverable state; the agent edits and tests freely.

cp_002 · restore 445–550ms*
Step 03

Held at the gate

Installing deps and creating the PR are real effects — so they wait. Nothing remote yet.

install · gh pr create — HELD
Step 04

Approved

A required check clears and a human approves what policy flagged.

j.okafor · security
Step 05

Executed + receipt

The broker runs the action, then emits a signed receipt to a GitHub Check.

receipt sha256:91fd…

The agent still completes the work. It just no longer installs dependencies or creates the PR silently — and the team defined which paths are automatic, what needs approval, and what is blocked.

* Restore figure from the ForkCell public benchmark. This walkthrough illustrates the controlled-run model, not a live production capture.

The proof object

Every action leaves an Action Receipt.

Not a log line you have to trust — a signed, independently verifiable record of what was allowed, under which policy, against which checkpoint, approved by whom. The audit trail writes itself.

  • Tamper-evident: hashed and chained
  • Maps the effect back to the policy decision that allowed it
  • Flows into the PR, ticket and audit chain
Action Receipt verified
actiongh pr create #1042
policy_hashsha256:7c1a…e9b4
checkpointcp_002 · restorable
decisionALLOW · gate passed
approverj.okafor · security
external_idgh:PR_kwDOA…7042
executed_at2026-04-18T09:21:07Z
receiptsha256:91fd…2a07
$ beforewire verify 91fd…2a07 → ✓ signature valid · chain intact
How teams roll it in

Start free and local. Grow into team governance.

01 · For developers
Available now

Start with ForkCell

The open-source governed execution cell — policy-bound runs, fast checkpoint/restore, receipts — is the part you can run today. Free and local.

pip install forkcell
02 · Controlled workflow
Early access

Attach one real entry point

Bind team policy, checkpoints, approval gates and receipts to one workflow. See what ran, what was blocked, what can be restored.

checkpoint adapteraction outbox
03 · AI Platform + Security
On the roadmap

Team governance buys in

Shared policy, approval routes, audit & evidence export, required checks — mapped onto your existing process. Receipts flow into PRs, tickets and audit chains.

What's live today: the ForkCell open-source core. The shared team control plane is in early access and rolling out — the runs shown above illustrate the model.

After BeforeWire

The person who has to say “yes” finally can.

Agents ship to prod — because Security signed off.

Policy is enforced at the wire, not promised in a doc. The blocker becomes the approver.

Every action is reversible or receipted.

Recoverable work checkpoints; irreversible work waits for a gate. No silent, unlocatable damage.

Audits read themselves.

Receipts are the evidence. Export the chain instead of reconstructing what an agent did from scattered logs.

Give agents production access — without giving up control.

Bring one action you'd never let an agent do unsupervised. We'll map it into a controlled run — policy decision, checkpoint, approval gate and a signed Action Receipt — so your security team can see exactly what they're saying yes to.

Book a demo View the coding run GitHub · ForkCell