Pick up the right work
Managed hooks capture useful findings and maintain task checkpoints. The next session receives relevant working context across Claude Code and Codex.
Choose how your agent picks up the work
Keep the progress.
Choose your next tool.
Your next session should start with the decisions you made, what you learned, and what comes next. OnlyBots brings that working context to Claude Code and Codex.
Start with free working memory. Add paid hosted operations when you need them. See pricing.
Why choose OnlyBots?
Inspect the handoff
Read saved memory, inspect a signed handoff, and record its receipt and outcome. You can check what was carried forward.
Try the workflow before paying
Hot memory is free. Paid retrieval and context operations use prepaid credits, with no subscription required.
Competitor pilot · two task families · October 2026
Matched task success.
Matched task success.
Lower API cost in our pilot.
OnlyBots, Cairntir and an update-only written handoff each passed all six trials. OnlyBots used 61.7% less model API spend than Cairntir and 15.6% less than the handoff across this pilot.
| Method | Passed | Input + output tokens | API cost |
|---|---|---|---|
| OnlyBots · local hot memory | 6/6 | 506,165 | $0.59 |
| Update-only HANDOFF.md | 6/6 | 561,678 | $0.69 |
| Cairntir 1.12.3 | 6/6 | 1,561,462 | $1.53 |
Both task families ran on GPT-6.1 Sol and Claude Opus 5.5 in fresh native threads. One family also ran in both directions between Codex and Claude Code. Cairntir used its official MCP integration, unmodified host policy and default embeddings; successful memory reads and writes were verified in every accepted trial. OnlyBots used its real hooks and free local hot memory.
One repeat on two synthetic development families, not an independent holdout or a general ranking. The handoff was cheaper in one of six cases. Costs are usage-based API estimates and exclude service fees, subscriptions, local compute and operator time. Setup and transport failures are excluded from this comparison and retained in campaign accounting. Native memory and the other named products have not been tested in this pilot.
Download all trial results, costs and source hashes · Cairntir version tested
Choose for the way you work
Several tools offer automatic capture and cross-tool memory. The useful comparison is how each fits your workflow, where your data lives, and what actually reaches the next session.
| Option | Workflow to explore | Documented approach |
|---|---|---|
| OnlyBots | Continue a task in your next Claude Code or Codex session | Managed hooks capture findings and task checkpoints, then deliver working context to fresh sessions. Free local hot memory, optional paid hosted operations, and inspectable signed handoffs. OnlyBots documentation · Measured results |
| Native memory and resume | Continue inside your existing coding tool | Built-in project memory and conversation history. Resume is a strong option when the original thread is available. Claude Code docs · Codex docs |
| claude-mem | Search and revisit your agent’s work | Automatic session capture, a local store, and progressive retrieval that fetches detail after finding relevant observations. Project documentation |
| Supermemory | Share memory across coding tools | Claude Code and Codex integrations provide automatic capture and recall, with personal and project scopes and hosted or self-hosted backends. Codex integration · Claude Code integration |
| Cognee | Connect sessions to a knowledge graph | Lifecycle capture and recall, with a shared dataset for the Claude Code and Codex plugins and local or remote operation. Integration documentation |
| Mem0 | Manage durable knowledge and session state | Its Claude Code integration combines memory tools and lifecycle hooks, including session-state preservation before compaction. Claude Code integration |
| Hindsight | Build on retained facts and reflection | A memory backend with coding-agent integrations, project and team banks, and published coding-task cost and correction measurements. Codex integration · Published measurements |
| Cairntir | Keep a portable local project history | A local store of decisions, unfinished work and outcomes, with budgeted handoffs and source provenance across supported hosts. Project documentation |
Sources checked October 20, 2026. Descriptions reflect the linked project documentation; this table is not a head-to-head performance ranking. Features and plans change. Check the linked source for current requirements.
What we have measured
In our October regression suite, OnlyBots and written handoffs completed 48 of 48 trials on each model. Cold restarts completed 38 of 48. OnlyBots’ model API cost was within 3% of cold restarts in either direction.
Across Claude Code and Codex, OnlyBots and written handoffs completed all 24 trials in the smaller cross-runtime suite. Native resume was cheaper than OnlyBots in the same-runtime comparison.
These are synthetic development tasks using local hot memory: 24 tasks per model in the main suite and six tasks in the cross-runtime suite, with two repeats. Costs exclude OnlyBots service fees. These larger suites compare continuation methods; the separate Cairntir pilot above is our first named-product comparison. Read costs, controls and limitations.
Try a handoff that matters to you
- Install OnlyBots and finish a small investigation in Claude Code or Codex.
- Inspect the saved finding with
onlybotsctl memory hot --scope all. - Start a fresh session in either tool and continue the task without repeating the finding.
- Check the result against your original requirement.
A useful test includes a correction or a decision that the repository alone cannot explain. Verify the resulting behavior as well as the saved text.