Skip to main content

Documentation Index

Fetch the complete documentation index at: https://kode-f177b001.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

kode plan analyzes your codebase and produces a compact, task-focused context graph — only the files and symbols that are actually relevant to what you’re trying to do. It uses Tree-sitter for AST-level parsing, so it understands your code structurally rather than relying on fuzzy string matching. The graph is token-capped (8,000 by default) to keep LLM costs predictable and context quality high.

​
Synopsis

kode plan <task>

​
What it does

kode plan takes your task description, parses your codebase with Tree-sitter, builds a dependency graph of files and symbols, and trims it to fit within the token budget. The result shows you exactly which files and symbols Kode thinks are relevant, how they relate to each other, and what the LLM will see when you run kode generate or kode loop. By default, kode plan prints a human-readable summary to the terminal. Use --packet to emit machine-readable JSON that you can pipe directly into kode generate or kode loop via --context-file.

​
Flags

​
--graph
boolean
default:"false"
Output the raw dependency graph as JSON instead of the human-readable summary. Useful for debugging context quality or piping into other tools.
kode plan --graph "fix the user service" | jq '.nodes | length'
​
--packet
boolean
default:"false"
Output the LLM-ready context packet as JSON. This is the compact representation that kode generate and kode loop consume via --context-file.
kode plan --packet "add authentication middleware" > context.json
​
--max-tokens
integer
default:"8000"
Maximum token budget for the context graph. Kode scores nodes by relevance and trims the graph from the bottom up until it fits within this limit.
kode plan --max-tokens 16000 "large cross-cutting refactor"

​
Examples

# Human-readable summary of the context graph
kode plan "add authentication middleware"

# Emit a context packet and save it for later
kode plan --packet "refactor auth" > context.json

# Count how many nodes are in the dependency graph
kode plan --graph "fix the user service" | jq '.nodes | length'

# Expand the context budget for a large task
kode plan --max-tokens 16000 "migrate database layer to new schema"

​
Typical workflow: plan then loop

Running kode plan --packet before a generation command is the most reliable way to handle multi-file tasks. The packet gives the LLM precise, AST-grounded context instead of a raw directory listing. 
# Step 1: build the context packet
kode plan --packet "add pagination to user list endpoint" > ctx.json

# Step 2: run the full pipeline with surgical context
kode loop --context-file ctx.json "add pagination to user list endpoint"
You can do the same with kode run: 
kode plan --packet "add input validation" > ctx.json
kode run --context-file ctx.json "add input validation to CreateUser handler"

​
Human-readable output

When run without --graph or --packet, kode plan prints a summary like this: 
── Context Graph Summary ────────────────────────────
  ✓  Nodes: 14  |  Edges: 22  |  Tokens: 3841 / 8000

── Files ────────────────────────────────────────────
  internal/user/handler.go
  ├─ function: CreateUser [high]
  ├─ function: validateUserInput [high]
  internal/user/service.go
  ├─ function: CreateUserService [medium]
  internal/middleware/auth.go
  ├─ function: AuthMiddleware [high]

── Plan Steps ───────────────────────────────────────
  Step 1: Add middleware to router
    internal/router/router.go
  Step 2: Implement auth validation logic
    internal/middleware/auth.go
Use --max-tokens to expand the context for large tasks (for example --max-tokens 16000). Larger contexts improve recall on cross-cutting changes but cost proportionally more tokens. For most single-feature tasks, the default 8,000-token budget is sufficient.