Navigate a large codebase without re-reading it every session.

On a 400k-line codebase, an agent spends the first several thousand tokens of every session re-discovering what it learned yesterday: where the payment logic lives, what calls what, which paths are off-limits. Then it edits a function without seeing that four other modules depend on it.

This is the case CTXone and AgentStateDeveloper (ASD) solve together. CTXone remembers what your team decided; ASD knows what the code actually does. An agent with both edits like someone who already knows the codebase.

Two kinds of context, one hub

• CTXone — remembered context. "The payment path is frozen until the audit clears." "We migrated off the legacy queue in Q1." Ranked, branchable, and attributable.
• ASD — structural context. Symbols, call graphs, effects, and ledger decisions, indexed and queryable in milliseconds through the same MCP connection.

What the agent does

recall("payment processing")          # CTXone: what did we decide?
  → [pinned] payment path frozen until audit clears

code_search("charge_card")            # ASD: where is it?
callers_of("payments.charge_card")    # ASD: blast radius
  → 4 call sites: billing, retries, admin, webhook

why-did-we("freeze payment path")     # CTXone: trace the decision
  → blame: security review, 2026-07-30, "SOC 2 evidence"

The agent now knows where the code is, what depends on it, and why it's off-limits — before it changes a line.

Setup is one flag

# Index the repo once
asd init && asd index .

# Start the hub with ASD wired in
ctx serve --http --asd-repo .

# ctx bootstrap will even offer to set ASD up for you

What changes

• Onboarding cost collapses. Structure is queried, not re-read. The agent asks ASD instead of paging whole files into context.
• Fewer blind edits. callers_of turns "I think this is safe" into a checked blast radius.
• Decisions survive the code. Frozen paths, deprecations, and gotchas live in memory next to the symbols they govern.

Next: ASD integration guide  ·  The partnership, in depth  ·  All use cases