- Auto model routing is now available end-to-end across the Editor (v1.3.2), CLI (v1.1.1), and Agent Manager (v1.1.3).
- Usage is billed strictly based on the specific model routed per turn, with full transparency via the new route log and usage cards.
- Mandatory plan approval checkpoints remain enforced under Auto mode; chat code auto-apply has been disabled to prevent unprompted code overwrites.
- A permanent 14-day promo trial has replaced the referral system.
The Technical Challenges of Model Routing
Developers want the efficiency of multi-model routing. It makes sense to use smaller, faster models for simple tasks (like file search or basic linting fixes) and reserve frontier models for complex architectural reasoning. However, relying on an automated router introduces inherent technical challenges for senior engineers: maintaining cost visibility and preserving execution control.
When an automated system routes a prompt, tracking which model executed the task—and verifying the associated cost—can become difficult. Without clear logging, auditing token usage across different model tiers requires parsing complex, disjointed telemetry.
Furthermore, autonomous coding agents require strict operational boundaries. If you are using a chat interface to ask questions about your codebase, you need a deterministic checkpoint before the agent modifies files. Routing logic should never override user consent.
Over the past week, MonkeysCode shipped a series of updates across the MonkeysCode Editor, CLI, and Agent Manager to resolve these routing challenges. Auto mode now provides per-model pricing transparency, retains mandatory plan approval, and strictly prevents chat code auto-apply.
End-to-End Auto Routing with Transparent Pricing
With the releases of Editor v1.3.1 and CLI v1.1.1, Auto mode is fully integrated into the standard workflow. We have removed the dedicated /frontier command and the Auto frontier button in the editor. Instead, Auto is now a standard option in the model selection menu.
In the CLI, selecting Auto is straightforward:
# Select Auto mode in the CLI
monkeyscode config set model auto
# Verify your current model selection
monkeyscode config get model
The most significant change is how the Context Kernel handles billing for Auto turns. Instead of applying a flat rate for the routing capability, the Context Kernel prices Auto turns exactly as the routed model. If the router assigns a task to a smaller open model, you pay the rate for that specific model. If it routes to a frontier model, you pay the frontier rate.
To make this verifiable, we shipped the GET /v1/usage/auto-summary endpoint. This endpoint powers the new Auto mode page and usage card in the web dashboard, surfacing a detailed route log. You can now see exactly which model was selected for each turn of your session and the corresponding cost. We also updated the platform to keep the Auto config partial index intact across schema:update operations, ensuring route logging remains consistent during database migrations.
Retaining Developer Control with Plan Approval
Routing efficiency cannot come at the expense of safety. When Capuchin (our built-in coding agent) formulates a plan, developers must have the opportunity to review it before execution begins.
In Editor v1.3.2 and Agent Manager v1.1.2, we shipped fixes to ensure that mandatory plan approval works seamlessly under Auto mode. The Model Proxy now explicitly enforces user plan approval and follows the admin profile strictly when routing.
We also overhauled the Editor UI to reflect these states accurately. The update introduces an Auto status strip, a per-tab change tray, and a single Auto route chip, giving you immediate visual feedback on which model is active and what changes are pending.
Crucially, we shipped a fix to prevent chat code auto-apply. Code generated during a chat session will remain in the chat interface until you explicitly apply it. This prevents the agent from overwriting code unprompted while you are simply exploring solutions or asking architectural questions.
Infrastructure, CI, and Cross-Platform Fixes
Alongside the Auto mode rollout, we shipped several updates to improve cross-platform stability and CI pipeline reliability for the Agent Manager and CLI.
For Windows users, we fixed an issue with MCP (Model Context Protocol) trust canonicalization and resolved a test failure caused by CRLF line endings. CLI test portability has been improved across all supported operating systems, and we updated the CLI terminal UI to support tabs, mouse interactions, and command menus.
In our CI pipelines, we resolved clippy errors and Rust formatting failures by enforcing cargo fmt in the rust-quality gate. We also fixed eslint errors in the Cloud/Workspaces/CloudJobService/windowDrag component. To ensure accurate testing, the CI workflow now explicitly builds the Context Kernel before running npm ci in the js-quality job and before the freshness check in agent-manager-quality.
Finally, Agent Manager v1.1.3 includes a performance audit, improvements to the research/docs/files workflow, a fix for turn-limit constraints, and better handling of billing payment failures. Agent Manager SSH support has also been merged.
Platform Updates: Documentation and Trials
On the web platform, we have transitioned to a permanent 14-day promo trial and removed the referral system entirely. This simplifies onboarding for new users evaluating the platform.
We also corrected routing for our documentation. Links pointing to docs.monkeyscode.com now correctly resolve to monkeyscode.com/docs, and we shipped a fix to the web dashboard to ensure "Getting started" links point to the right destination. Additionally, the documentation engine now correctly renders Markdown files that contain placeholders in code blocks.
Takeaway
By integrating Auto routing directly into the core workflow and pricing it based strictly on the routed model, MonkeysCode ensures developers get multi-model efficiency without sacrificing visibility or control. With mandatory plan approval and detailed route logs now standard, you can leverage the speed of smaller models and the reasoning of frontier models with complete operational transparency. Review your usage logs today in the MonkeysCode Dashboard.
Top comments (0)