A new AI coding agent dropped on Hacker News this week, and it's got the kind of feature list that makes you wonder if they're overcompensating for something. Kodiqa Agent launched June 6 with support for seven different API providers—local Ollama (free), Claude, OpenAI, DeepSeek, Groq, Mistral, and Qwen—all accessible through a single unified interface. The pitch: one agent, every model, zero limits.
Built Different From Day One
Kodiqa isn't just another wrapper around an LLM API. The project ships with 78 slash commands and 26 built-in tools covering file operations, git workflows, web search, memory management, clipboard integration, multi-edit capabilities with undo (up to 10 levels per file), diff application, and LSP-based project indexing. Installation is straightforward—pip install kodiqa—and it's Python under the hood with an AGPL-3.0 license. The test suite alone runs 473 pytest cases. The standout differentiator might be lazy MCP tool loading. Most agents inject every connected MCP server's JSON schema into every API request, which adds up fast when you're running large tool servers. Kodiqa exposes just three meta-tools—mcp_search, mcp_tool_schema, and mcp_call—and lets the model discover tools on demand. According to their docs, this approach cuts per-turn token costs by roughly 94% compared to schema injection.
The Multi-Model Angle
Beyond single-provider mode, Kodiqa offers multi-model consensus querying—hit all configured models simultaneously and merge the best answers. There's also cross-provider failover: if your primary provider hits rate limits or goes down, the system automatically retries on the next configured provider without losing context. That's a feature power users have been asking for since the Claude API outage last year. Permission modes come in three flavors—default (arrow-key confirm everything), relaxed (auto-approve file ops), and auto (no confirmations). Plan mode lets the AI explore and plan before implementing, which is useful when you want visibility into what a refactoring operation will actually touch. Batch edit review queues up changes for per-file accept/reject with arrow key navigation.
How It Stacks Up
The comparison table on their GitHub README puts Kodiqa against Claude Code ($20/month or pay-per-token), Aider (pay-per-token only), Gemini CLI (free but single provider), and OpenCode (75+ providers but no consensus mode). Kodiqa claims unique advantages in free local models, multi-model consensus, custom plugins, sub-agents, conversation branching, budget limits via /budget, auto-lint with ruff integration, and automatic model discovery from APIs.
Developer Experience Details
On the UX side: arrow-key navigation for all prompts (no letter shortcuts), compact streaming that hides code blocks during generation to show progress instead (/verbose toggles full output), stream interrupt with Esc or Ctrl+C, animated spinner for thinking/reasoning blocks, and five color themes. Session management includes auto-save on quit, conversation recovery on crash, checkpoint save/restore, export to markdown, and branching via /branch.
The Bottom Line
Kodiqa is trying to be the terminal-based equivalent of a universal remote—convenient if you actually use multiple providers, overkill if you're happy with Claude Code or OpenCode. But the lazy MCP loading alone justifies attention from anyone running heavy tool integrations. Worth installing and keeping around as a backup agent.