Now in public beta

Capability Broker
for AI Agents

Policy engine, approval flows, and audit trail for any agent runtime. Give your agents the right tools — and nothing more.

AI agents can call any tool, access any file, and hit any API. Without a governance layer, that's a production incident waiting to happen. Outcall sits between your agent and the outside world — enforcing policy, capturing every action, and routing dangerous calls to a human approver.

How Outcall works

Every tool call flows through a policy evaluation pipeline.

Agent
Tool Request
bash("rm -rf")
Guardrails
Pattern Check
5 rules
Policy
Rule Eval
allow / deny
Execute
Deny
Ask Human

5 built-in guardrails

Deny-by-default protection against the most dangerous agent behaviors.

git push --force
Overwrites remote history
deny
branch: main (protected)
Direct push to protected branch
ask
shell=True subprocess
Shell injection vector
deny
../../../etc/passwd
Path traversal attempt
deny
env: AWS_SECRET_KEY
Secret exfiltration risk
ask

Everything you need to govern agents

From policy evaluation to human approval to immutable audit logs.

Policy Engine

Declarative allow/deny rules per agent, tool, and environment. YAML or JSON configuration.

Approval Gates

Human-in-the-loop approvals via Web UI, Telegram, or Discord. Configurable timeouts and escalation.

Audit Trail

Immutable log of every agent action: tool called, policy decision, approval status, and timestamp.

Tool Marketplace

8 built-in executors. Compose permissions per executor. Author custom executors with typed manifests.

Human approval,
wherever you work

When an agent wants to do something risky, Outcall routes the approval request to the right channel — instantly. Approve or deny from your preferred interface.

Web UIVisual dashboard with full context
Pro
TelegramApprove from any device
Pro
DiscordTeam approvals in your server
Pro
Approval required2s ago
Agent
clawie-dev
Requested action
git push origin main
Policy match
protected-branch: main requires approval

8 built-in executors

Every executor has a typed permission manifest. Compose exactly what each agent needs.

>_
bash
Shell execution
F
file-read
File system read
W
file-write
File system write
H
http
HTTP requests
B
browser
Browser automation
D
db-query
Database queries
G
git
Git operations
P
process
Process management

Works with any agent runtime

ClawieLangChainOpenAI AgentsCustom Runtimes

Start governing your agents

Free core tier includes the CLI, 5 guardrails, and deny-by-default policy. No credit card required.