Context & Memory

jcode automatically understands your project and provides the agent with rich context. You can also customize behavior through AGENTS.md files.

Automatic Context

When jcode starts, it detects and provides to the agent:

Context	How It’s Detected
Working directory	Current directory
Platform	OS and architecture
Git branch	git rev-parse --abbrev-ref HEAD
Git status	Uncommitted changes (dirty/clean)
Last commit	git log -1
Project type	Detected from files like go.mod, package.json, Cargo.toml, etc.
Directory tree	Shallow tree (depth 2, max 200 lines)
Environment	Local or SSH, with labels
Skills	Available skill packs and slash commands

AGENTS.md — Custom Instructions

Use AGENTS.md files to give the agent project-specific instructions. These are loaded automatically and injected into the agent’s context.

Where to Place AGENTS.md

Three levels, all merged together:

Priority	File	Scope
1 (lowest)	~/.jcode/AGENTS.md	Global — applies to all projects
2	./AGENTS.md	Project-level — shared with team
3 (highest)	./AGENTS.local.md	Personal — typically git-ignored

Example AGENTS.md

# Project Conventions

- Use Go 1.22+ features (range over int, etc.)
- Follow the existing error wrapping pattern with `fmt.Errorf`
- All public functions must have doc comments
- Run `make lint` before considering any task complete

# Architecture

- `cmd/jcode/` — Entry points
- `internal/` — All internal packages
- `web/` — Vue 3 frontend

# Testing

- Run tests with `make test`
- Integration tests are in `tests/` directory

Include Directives

AGENTS.md supports @include directives to reference other files:

# Project Conventions

@include ./docs/conventions/coding-style.md
@include ./docs/conventions/testing.md

• Includes are resolved relative to the file’s directory
• Recursion depth limited to 5 levels
• Circular includes are detected and blocked
• Total content is capped at 40,000 characters

Context Compaction

When the conversation gets long, jcode automatically summarizes older messages while keeping recent ones intact.

Auto-Compaction

Configure when compaction triggers:

{
  "compaction": {
    "enabled": true,
    "threshold": 0.75,
    "keep_recent": 6,
    "summary_model": "openai/gpt-4o-mini"
  }
}

Setting	Default	Description
enabled	—	Enable auto-compaction
threshold	0.75	Trigger compaction when context reaches this fraction
keep_recent	6	Number of recent messages to preserve
summary_model	—	Model to use for generating summaries

Manual Compaction

Type /compact in the TUI to trigger compaction at any time.

Smart Prompt Caching

Enable prompt caching to reduce redundant computation across turns:

{
  "prompt": {
    "cache_enabled": true
  }
}