Llamatik Code on Llamatik Documentation

Getting Started

Mon, 01 Jan 0001 00:00:00 +0000

Getting Started with Llamatik Code#

Requirements#

IntelliJ IDEA 2026.1 or any JetBrains IDE built on platform 252 or later
A compatible GGUF language model file on disk
macOS, Linux, or Windows with 64-bit JDK 21+

Installation#

Open Settings → Plugins → Marketplace.
Search for Llamatik Code.
Click Install, then restart the IDE.

Alternatively, download the .zip from the JetBrains Plugin Marketplace and install via Settings → Plugins → Install Plugin from Disk.

Agent & Tools

Mon, 01 Jan 0001 00:00:00 +0000

Agent & Tools#

The Llamatik Code agent is a tool-driven loop that takes a natural-language instruction, plans steps, calls built-in tools, and applies changes to your project with a safe diff preview.

How the agent works#

You type an instruction in Agent mode.
The agent makes a first LLM pass to produce a plan.
It executes each step by calling one or more tools.
Results are fed back into the loop until the task is complete.
File writes are shown as a diff preview — you approve before changes land on disk.
After apply, the agent optionally triggers a build and interprets errors.

Built-in tools (available since 1.2.0)#

Tool	Description
`read_file`	Read the content of any project file
`write_file`	Write or overwrite a file (always previewed)
`list_files`	List files in a directory
`search_code`	Full-text search across the project
`run_terminal`	Execute a shell command
`get_diagnostics`	Fetch current IDE inspection/compiler errors
`explain_build_failure`	Interpret the last build failure
`run_diagnostics`	Run plugin self-diagnostics

Git tools (1.3.0+)#

Tool	Description
`git_diff`	Diff of current changes (staged + unstaged)
`git_status`	Working-tree status
`git_log`	Recent commit history
`git_blame`	Line-level blame for a file
`git_changed_files`	Files changed in the current diff

Project intelligence tools (1.3.0+)#

Tool	Description
`index_search_files`	Fast file search by name pattern
`index_search_symbols`	Symbol search across the project index
`index_list_recent`	Recently opened or modified files
`psi_find_usages`	Find all usages of a symbol via PSI
`psi_get_type`	Resolve the type of an expression
`psi_get_declarations`	List declarations in a file
`smart_search`	Fuzzy search ranked by recency and relevance

Codebase rules (1.3.0+)#

Write persistent natural-language rules that the agent follows in every session:

MCP Servers

Mon, 01 Jan 0001 00:00:00 +0000

MCP Servers#

Llamatik Code is an IDE-native MCP (Model Context Protocol) host. Connect any MCP-compatible tool server and let the agent call its tools through the same pipeline as built-in tools.

Available since version 1.4.0.

Adding a server#

Open Settings → Llamatik Code → MCP Servers.
Click Add server.
Fill in the server details:

Field	Description
Name	A label for the server (e.g., `github`, `filesystem`)
Command	The executable to run (e.g., `npx`, `uvx`, `python`)
Arguments	One per line (e.g., `-y @modelcontextprotocol/server-github`)
Environment variables	`KEY=VALUE` pairs for secrets or config
Working directory	Optional; defaults to the project root
Auto-start	Start automatically when the project opens
Trust level	LOW / MEDIUM / HIGH — affects default approval behaviour

Click Test connection to verify the server starts and lists its tools.

Managing servers#

Enable / disable — toggle a server without removing its configuration.
Refresh tools — re-discovers tools from a running server after an update.
View logs — shows the last 1000 lines of stderr. Useful for debugging.
Edit — modify any field; changes take effect on the next connection.
Remove — stops the server and removes the configuration.

How MCP tools appear in the agent#

Discovered MCP tools are bridged as first-class agent tools under mcp_<serverName>_<toolName>. They appear in the agent trace like built-in tools.

External Knowledge

Mon, 01 Jan 0001 00:00:00 +0000

External Knowledge#

External Knowledge gives the Llamatik Code agent the ability to search the web, read documentation pages, research errors, and query private docs sources — all from inside the IDE, with citations.

Available since version 1.5.0.

Enabling it#

Open Settings → Llamatik Code → External Knowledge.
Toggle Enable external knowledge.

Web search#

The agent searches the web via DuckDuckGo — no API key required.

Tool	Description
`external_search`	General web search
`external_search_official_docs`	Search restricted to official documentation sites
`external_search_github_issues`	Search GitHub Issues for known bugs and solutions

	Free	Pro
Searches per day	5	50
Results per search	5	10

Page reading#

Tool	Description
`external_read_page`	Fetch and read a single URL
`external_read_pages`	Read multiple URLs in parallel

	Free	Pro
Pages per task	3	10

Research orchestration#

The research orchestrator detects the intent behind a query and plans a multi-step research workflow automatically.

Code Health Monitor

Mon, 01 Jan 0001 00:00:00 +0000

Code Health Monitor#

The Code Health Monitor is a proactive background analysis layer that continuously inspects your codebase for structural problems, regression risks, and refactoring opportunities — without requiring a model call for every insight.

Available since version 1.5.0.

Enabling it#

Open Settings → Llamatik Code → Code Health Monitor.
Toggle Enable Code Health Monitor.
Choose an analysis mode and run trigger.

Analysis modes#

Mode	Files per scan	Max runtime	Model calls/hour	Who
`LIGHT`	5	8 s	3	Free + Pro
`BALANCED`	30	30 s	20	Pro only
`AGGRESSIVE`	150	120 s	50	Pro only

Free users are always capped at LIGHT mode regardless of settings.

Settings Reference

Mon, 01 Jan 0001 00:00:00 +0000

Settings Reference#

All Llamatik Code settings live under Settings → Llamatik Code.

Llamatik Code (main page)#

Setting	Description
License key	Pro license key
Trial status	Days remaining in free trial

Inference#

Setting	Description	Default
Model path	Absolute path to a GGUF model file	—
Context size	Token context window	4096
Threads	CPU threads for inference	System cores / 2
GPU layers	Layers offloaded to GPU (0 = CPU only)	0
Temperature	Sampling temperature	0.7
Top-p	Nucleus sampling threshold	0.9
Agent fallback policy	ABORT / RETRY / ESCALATE on step failure	ABORT
Run benchmark	Measure tokens/s for the current model	—

MCP Servers (1.4.0+)#

Setting	Description
Server list	All configured MCP servers
Add server	Open the server editor
Auto-approval rules	Per-tool and per-server rules (Pro)

Per-server fields: name, command, arguments, environment variables, working directory, enabled, auto-start, trust level.

Changelog

Mon, 01 Jan 0001 00:00:00 +0000

Changelog#

1.5.0 — 2026-05-08#

Code Health Monitor#

New opt-in background analysis layer for proactive code quality monitoring.
Static checks without a model: large files (>300 lines), TODO/FIXME accumulation (≥3), empty catch blocks, deprecated API usage.
Regression risk analyser for git diffs: removed public APIs (HIGH), deleted tests (HIGH), Gradle dependency changes, removed error handling, AndroidManifest modifications.
Before-commit assistant: summarises staged changes and flags regression risks.
Watched areas: pin files, directories, packages, or modules for priority analysis (Free: 1, Pro: 20).
Refactor opportunity detection: long functions (>40 lines), repeated catch patterns.
Session context tracking: recent files, current branch, last build failure.
14 new agent tools for all Code Health operations.
New settings page: Settings → Llamatik Code → Code Health Monitor.
Free: LIGHT mode, manual trigger only. Pro: BALANCED (30 files/30s) and AGGRESSIVE (150 files/120s) modes with background scheduling.

External Knowledge#

Agent can search the web, read docs pages, and research errors from inside the IDE.
Research orchestrator: detects intent (library docs, bug, API, migration, version) and plans multi-step searches.
Smart source routing to kotlinlang.org, developer.android.com, developer.apple.com, spring.io, github.com based on query context.
Error research: detect build tool and framework; produce structured fix plans.
Private docs indexing: add up to 10 (Pro) or 1 (Free) custom documentation sources.
Citations with every external research result (source type, confidence, URL).
16 new agent tools for all External Knowledge operations.
New editor action: Search Docs & Web in the right-click context menu.
New settings page: Settings → Llamatik Code → External Knowledge.
Free: 5 searches/day, 3 pages/task. Pro: 50 searches/day, 10 pages/task.

1.4.0 — 2026-05-07#

MCP Server Support#

Llamatik Code is now an IDE-native MCP host.
New settings page: Settings → Llamatik Code → MCP Servers.
Full server lifecycle: add, edit, remove, enable/disable, test connection, view logs, refresh tools.
6 new built-in agent tools: mcp_list_servers, mcp_list_tools, mcp_call_tool, mcp_list_resources, mcp_read_resource, mcp_list_prompts.
Discovered MCP tools bridged as mcp_<server>_<tool> first-class agent tools.
Approval dialog with LOW / MEDIUM / HIGH risk level on every MCP tool call.
HIGH-risk tools never auto-approved.
Free: 1 active server, manual approval. Pro: unlimited servers, per-tool and per-server auto-approval.
Secrets redacted from all log and trace output.

1.3.0 — 2026-05-06#

Agent tools#

Git: git_diff, git_status, git_log, git_blame, git_changed_files.
Project index: index_search_files, index_search_symbols, index_list_recent.
PSI: psi_find_usages, psi_get_type, psi_get_declarations.
Codebase rules: rules_get, rules_set, rules_clear.
Run configs: run_config_list, run_config_execute.
KMP native: kmp_diagnose, kmp_fix_plan.

Aggressive Mode (Pro)#

AGGRESSIVE execution mode for decisive agent runs.
Configurable fallback policy: ABORT / RETRY / ESCALATE.

Model Intelligence#

ModelCapabilityService: detects tool use, long context, code, and instruction-following capability.
ModelBenchmarkService: measures tokens/second and first-token latency.
SmartSearchService: fuzzy, recency-ranked file and symbol search.

Snapshots & Context Recovery#

ProjectSnapshotService: lightweight project state snapshots before agent runs.
ContextRecoveryService: detects context drift and offers recovery options.

1.2.0 — 2026-05-04#

Core agent#

Tool-driven agent loop with two LLM passes (plan + execute).
Safe diff preview before any file write.
Build verification after apply.
Agent trace panel with tool calls, duration, and results.

Multi-task agent (Pro)#

Parallel task execution with a live task queue panel.
Task cancellation and live status tracking.

Chat sessions#

Persistent multi-session chat store with per-project isolation.
Chat session list panel with last-message preview.
Parallel Ask-mode conversations.

Context actions#

Editor and project-view right-click actions: Ask About Selection, Explain This File, Refactor Selection, Fix Selected Code, Generate Tests, Add Documentation, Edit with Prompt, Explain Failing Build, Run Diagnostics, Explain Current Git Changes.