Memory & Context

Re-verify stored agent memory against the live environment and flag stale entries

Build a knowledge graph of team concepts, terms and the links between them

Search past AI coding sessions for settled decisions, banned tools and merge checks

Search your past AI coding sessions with the deja CLI for naming rules and checks

Add persistent memory and filesystem access to Deep Agents with storage backends

Persist LangGraph state with checkpointers, thread IDs, time travel and stores

Find and read playbooks and lessons Autocontext produced for a scenario

Read the last run's record and save a short record of this run for the next one

Search past chat sessions for original messages when memory alone is not enough

Store important step outcomes in long-term memory with automatic forgetting

Save step context to agent memory without moving or speaking by default

Manage context window budgets and session handoffs between agents

Propose harness improvements from learned lessons and apply or roll them back

Add code annotations to a codebase so AI tools keep better context

Read long web pages or knowledge bases in segments and store key points as you go

Cut long conversation transcripts to a fixed size by choosing which turns to keep

Rank retrieved passages, drop irrelevant ones and flag hidden prompt injection

Write verifiable agent track records and check the risk of agents and tokens

Index and search local coding agent history across Codex, Gemini, Cursor and more

Cut LLM API costs and latency with exact, semantic and provider prompt caching

Tag, annotate and later find which agent session handled a given task

Export and import agent sessions to hand off work context to a colleague

Search past agent sessions and trace code lines back to the session that wrote them

Check how much of the context window and token budget is in use

Start a fresh session with clear context when the current one grows too large

Keep memory across sessions in tiered markdown files with sync and consolidation

Audit an agent's context layout for caching, token spend and CLAUDE.md structure

Ask source-first questions about libraries and local resources with btca

Answer library and framework questions from local sources through btca

Write spec files in compressed caveman shorthand to cut token use by about 75 percent

Cut response length by 30 to 50 percent to save context and tokens

Measure Claude Code prompt-cache hit rates and find sessions with poor reuse

Import Claude Code and Codex history and move dashboard data between machines

Review CLAUDE.md and auto-memory files for stale or oversized facts

Suggest cost, speed and error fixes for Claude Code usage from past sessions

Search a Claude Code session transcript for a pattern and show surrounding messages

Replay a Claude Code session turn by turn with a short summary of each message

Record verified project facts, lessons and decisions for later sessions

Store small validated lessons that load automatically into every later session

Restore a saved business-state snapshot to continue exactly where you left off

Save confirmed conclusions and open questions so the next session can resume

Publish a redacted local coding session to a read-only Beam catalog

Write a clipboard-ready handoff prompt so another agent can continue the task

Turn Codex, Claude Code or OpenClaw session logs into a searchable single HTML file

Hand local work to a shared OpenClaw Gateway session and get its URL back

Write experiment, infra or failure notes that future agents can act on

Restructure and deduplicate a shared notes folder and refresh its index

Track complex tasks in persistent markdown plan, progress and notes files

Write a traceable handoff note that transfers an existing research project