Free AI Agent - No API Key on Device
agent.mona.expert is an open-source control plane that sits between an autonomous AI agent and the external tools, files and APIs it is permitted to touch. Rather than being another chat wrapper, it acts as a policy and enforcement runtime: it evaluates agent intent against a per-agent manifest and decides whether a tool call is allowed, requires confirmation, or is blocked. Each agent is configured with its own permitted tools, risk tier, tenant, approval rules and spending budget, so operators can put firm boundaries around otherwise free-running agents. The project positions itself around two concerns: security and cost control. On the security side it advertises prompt-injection and instruction-takeover pattern detection applied to prompts, retrieved content, tool outputs and encoded payloads, along with mTLS-based component identity, sandboxed execution and a credential vault so that API keys stay out of the agent runtime. On the cost side it provides per-agent token budgets and daily caps, tracks actual spend, and reports estimated savings from blocked or rejected calls that were never billed. Every decision is written to an append-only, SHA-256 hash-chained audit log that streams into a live console, letting an operator see which agent performed which action, the reason for the allow or deny decision, and whether the integrity chain still verifies. The homepage presents console views for fleet status, budgets, live logs, audit history and spend insights as real product surfaces once agents are connected. Pricing is listed as free, with a separate plans page; the exact license, public repository and self-hosting requirements are not stated on the homepage.
- Category
- AI Agents
- Pricing
- Free
Problem solved
Autonomous agents can call tools, read files, reach network APIs and consume tokens with little oversight, which creates prompt-injection exposure, credential leakage and unpredictable bills. agent.mona.expert adds a policy, identity, sandboxing and budget layer plus a tamper-evident audit trail, so operators can constrain what agents do and what they cost, and can prove afterwards what happened.
Use cases
- Enforcing per-agent allowlists and approval rules before tool calls execute
- Detecting and blocking prompt-injection patterns in prompts, retrieved content and tool output
- Setting per-agent daily token budgets and spend caps for LLM workloads
- Keeping API keys in a vault instead of inside agent code or configuration
- Producing an append-only, hash-chained audit trail of agent decisions for compliance review
- Monitoring a fleet of agents through a live dashboard with status, logs and spend insights
Audience
- Developers
- ML Engineers
- Enterprise
- Startups
Pros
- Policy, budget and audit controls are configured per agent rather than globally
- Hash-chained, append-only audit log with chain verification
- Credential vault design keeps API keys out of the agent runtime
- Free tier advertised with a hosted console for fleet visibility
- Addresses both security and cost governance in one layer
Cons
- Homepage is marketing-heavy; deep technical documentation is not exposed there
- License, public repository and self-hosting requirements are not specified on the homepage
- Security claims such as injection detection are not independently documented or benchmarked
- Operational setup effort implied by mTLS, sandboxing and vault integration
- No published release history, versioning or starred community repository surfaced
Official website: https://agent.mona.expert/
Canonical LambdaBase page: https://www.lambdabase.com/tools/free-ai-agent-no-api-key-on-device