The situation
Multiple agents share a project. Sessions get lost, experience is not reused and routine tasks need manual attention.
What comes out
Searchable history, reviewed handoffs, reusable memory and scheduled/loop execution records
Local agent workbench
LannerFind work. Resume it. Build memory. Let agents take the next step.
Lanner brings agent sessions, interactive terminal tasks and reviewed cross-agent handoffs into one local workbench. With explicit permission, it can build reusable memory, run scheduled and bounded loop tasks, and inspect agent behavior, model usage and cost.
LannerLocal agent workbench
The problem
Sessions and tasks are scattered across tools; prior work rarely informs the next run.
What gets easier
Find work, review a handoff, opt into memory and plan bounded follow-up tasks.
Who it is for
Developers, technical leads and teams using multiple AI agents
Desktop / Lanner
Find the work. Keep the context. Continue with agents.
Manage local sessions, tasks, reviewed handoffs and opt-in memory. Schedule routine work or use bounded loops with quality gates, then inspect behavior and usage.
Explore product ↗Find earlier decisions and next actions in authorized sessions.
Choose relevant memory and review context before proceeding.
Run a follow-up task or iterate within agreed limits and quality gates.
Illustrative workflow · Select a step to explore
A PRACTICAL EXAMPLE
What does this mean at work?
What goes in
Authorized local sessions, tasks, optional memory and providers
How to check it
Check resume, memory permissions, quality gates and usage
Discuss this use case ↗LANNER / AGENT WORKBENCH
Don't just recover a chat.
Pick up the actual work.
When several agents help you today, will you know what they did tomorrow? Lanner brings sessions, tasks and usage into a workbench on your own machine. Resume execution, opt into reusable memory, and schedule or loop through bounded tasks.
What did the agent finish yesterday?
How do I continue later—or hand context to another agent?
Can useful work become experience for the next task?
What can run on a schedule, and when should I approve?
ONE WORKBENCH
Seven views, one workbench
01Overview
Home
Review agent activity, pending memory reviews and authorized usage summaries.
02Find context
Sessions
Find and read local conversations across agents; filter, search and export.
03Resume work
Tasks
Use real interactive terminals and resume work after a browser refresh.
04Bounded autonomy
Loops
Iterate towards a goal with bounded rounds, quality gates and human approval.
05Scheduled work
Automations
Run planned tasks, inspect results and pause or retry where appropriate.
06Build experience
Memory
Explicitly authorize facts, preferences and reusable skills on a per-project basis.
07Understand usage
Agent analysis
Examine agent behavior alongside available model, token and cost metadata.
A concrete cross-agent handoff
Codex started it yesterday.
Need Claude Code to continue?
The current cross-agent handoff focuses on a controlled Codex-to-Claude Code flow. It is not universal session migration.
Discuss your agent workflow- 01
Find the session
Locate the previous task and review what is done and what remains.
- 02
Review the handoff
Compile a bounded, redacted packet of goals, decisions and next actions.
- 03
Start the next agent
Approve the handoff and launch a new session in an authorized workspace.
- 04
Keep useful experience
Optionally retain relevant memory and choose a scheduled or bounded loop task.
LOCAL FIRST
Local-first, with permission for every data scope.
Local browsing, usage collection and memory generation are separate choices. Viewing sessions does not enable background collection or authorize sending their text to a model.
Local browsing
Read and search authorized sessions on your own machine.
Usage metadata
When separately configured, collect timestamps, models and tokens without conversation text by default.
Models & memory
Opt-in generation and task runs may send selected content to the configured model provider.
More than a browser window
Use the web, a terminal, or a script.
Web workbench
Browse sessions, tasks, loops, automations, memory and analysis.
TUI
Connect to an existing Lanner instance from a terminal.
CLI / REST
Access structured status and controlled handoff workflows.
Lanner / CAPABILITIES
What it helps with
Session search, task resume and agent handoff
Find prior sessions and tasks
Opt-in memory, automations and bounded loops
Review agent handoff and opt-in memory
Agent analysis, model usage and local collection
Schedule bounded work and inspect analysis
WORK IN CONTEXT
How the work proceeds
Lanner / Agent workbench
How can an agent pick up
yesterday's work and learn from it?
Sessions and tasks are scattered across tools; prior work rarely informs the next run.
Your objective
Recover prior work, review context and plan a controlled next run.
- Find prior sessions and tasks
- Review agent handoff and opt-in memory
- Schedule bounded work and inspect analysis
LannerWORK / 01
Illustrative deliverable
Agent follow-up plan
- Previous work
- Sessions and next actions
- Reusable context
- Approved memory and handoff
- Next step
- Automation or bounded loop
Find work, review a handoff, opt into memory and plan bounded follow-up tasks.
An illustrative workflow using a prepared task.
DESIGNED WITH BOUNDARIES
Before connecting
Data and models
Local sources require authorization. Memory extraction or external models may transmit selected content.
No. Applications and infrastructure serve different needs; choose or combine only where your workflow calls for it.
EnterpriseYOUR NEXT CHAPTER
Start with one problem.
Solve it well.
Begin with one workflow and a result you can check.