Calliope AI IDE v1.5.0 is live. This release doubles down on what makes the IDE genuinely different from Cursor, Windsurf, and every other AI coding tool on the market: multi-agent workflows that actually scale to real engineering problems. Council Mode has been significantly upgraded, the AutoAgent backend is dramatically faster, and the staged changes review system is now precise enough that you will wonder how you worked without it.

Download Calliope AI IDE v1.5.0

What is Calliope AI IDE?

Calliope AI IDE is a VS Code-based AI development environment with five operating modes (Agent, Ask, Plan, Debug, Council), support for 9+ LLM providers, and multi-agent orchestration built in. Zero telemetry. No data collection. Bring your own keys.

What is new in v1.5.0

Council Mode: Smarter Deliberation, Better Decisions

Council Mode, where multiple agents with different backends deliberate before committing to a solution, has been significantly upgraded in this release.

Councils can now be configured with named roles. Assign one agent as Architect, one as Security Reviewer, one as Performance Analyst. Each role gets a tailored system prompt and can be backed by a different model. A senior engineer reviewing a Council session will see clearly structured position statements from each agent, a synthesis, and a final recommendation with dissent noted where it exists.

Councils also now support async deliberation. Kick off a Council session, keep coding, and get a notification when the agents have reached a conclusion. For large architectural decisions, this is a game changer.

AutoAgent Backend is 50% Faster

The AutoAgent backend has been completely rewritten for v1.5.0. Task decomposition is now parallel rather than sequential, meaning complex multi-step agent tasks complete in roughly half the time. For autonomous refactoring runs and full feature implementation tasks, the difference is immediately noticeable.

Staged Changes Review: Line-Level Precision

The staged changes review system now operates at line-level granularity rather than file-level. Every edit an agent proposes shows a tight, contextual diff with the surrounding code. You can accept or reject individual hunks, not just whole files. For large codebases where agents touch many files in a single session, this gives you exactly the control you need.

New: DeepSeek R2 and Grok 3 Support

v1.5.0 adds native support for DeepSeek R2 and xAI Grok 3 as LLM backends, bringing the total provider count to 11. DeepSeek R2 in particular is an excellent choice for Plan Mode on complex architectural tasks. Grok 3 handles rapid-fire Ask Mode queries at low cost.

Git Worktree Sandbox Improvements

The Git Worktree sandbox mode now properly handles monorepos with nested submodules. Previously, agents working in a worktree on a monorepo would occasionally lose track of submodule state. That is fixed. Worktree cleanup on session end is also more reliable.

Plan Mode: Approval Gates for External Calls

Plan Mode now supports configurable approval gates specifically for external API calls and shell commands. You can allow agents to read and write files freely while requiring explicit approval before any network request or shell execution. This is the right default for most production codebases.

Bug Fixes

• Fixed: top_p error with Claude models (reported by julatec, issue #5 ) – Users running Claude Opus 4.5 and other Anthropic models were hitting 400: top_p cannot be set to -1 for this model on every agent call. The IDE was passing an invalid default parameter that Anthropic rejects. Fixed in v1.5.0. Thanks to Jose for the detailed report including logs.
• Fixed Council Mode sometimes producing duplicate agent responses when using the Anthropic provider
• Fixed AutoAgent occasionally failing to restore working directory after a failed task
• Fixed Plan Mode not correctly resolving relative imports in TypeScript projects
• Resolved an issue where Debug Mode would lose breakpoint context after an agent-suggested edit
• Fixed the model picker not persisting selection across sessions for secondary agent backends
• Fixed a memory leak in long-running Ask Mode sessions with large codebases

Download

Download Calliope AI IDE v1.5.0

Available for macOS (Apple Silicon and Intel) and Windows. Free to use. Bring your own API keys for any supported provider.

What is Next

v1.6.0 will introduce persistent agent memory across sessions, a new Code Archaeology mode for understanding unfamiliar codebases, and tighter Zentinelle integration for enterprise compliance workflows.