Cline CLI v3.0.68 fixes agent-team slowdown that could grow teams.db to gigabytes
Cline CLI v3.0.68 fixes a performance bug in which every streamed chunk re-saved the entire agent-team state, slowing long runs and letting ~/.cline/data/db/teams.db grow to gigabytes. The release also refreshes the model catalog and changes default models for five providers.
Cline CLI v3.0.68 fixes a persistence bug that made agent-team sessions progressively slower and could inflate the local ~/.cline/data/db/teams.db database to gigabytes, according to the release notes. The release also refreshes the CLI's model catalog and changes the default model for five providers.
The release was published by the github-actions account on 02 Oct at 04:42 (the source page does not display the year) and corresponds to commit 241c188. It is tagged cli-v3.0.68 and marked Latest on the cline/cline repository.
Agent teams: the fix
Per the release notes, the previous behavior was that every streamed chunk caused the full team state to be re-saved. Two consequences followed:
- Team runs got slower the longer a session continued.
~/.cline/data/db/teams.dbcould grow to gigabytes.
v3.0.68 stops that pattern, so agent teams are described as staying fast in long sessions and while teammates are streaming. Existing team data is compacted automatically the first time the database is opened. Users with large teams.db files should see them shrink after first launch on the new version. The notes do not give benchmark figures for the speedup or typical post-compaction file sizes.
Model catalog refresh
The catalog refresh changes default models for five providers:
| Provider | New default model |
|---|---|
| DigitalOcean | GPT-6.1 Sol |
| GMI Cloud | Qwen 3.8 Flash |
| NanoGPT | Pareto 26.10 Preview |
| Nvidia | DeepSeek V4.1 Flash |
| Ofox | GPT-6.1 Sol |
The notes do not state which models these replace. Context window sizes, per-token pricing and benchmark results for the new defaults are not disclosed in the release. Model names are reproduced as listed in the changelog; Cline does not publish capability details for them here.
This is a tooling release, not a model release. Cline is not launching or training any of these models. It is changing which model each provider entry preselects.
Upgrade notes
- Compaction of existing team data runs automatically on first open, so no manual migration is described.
- Default-model changes apply to provider defaults. Whether users with an explicitly saved model selection are affected is not specified in the notes.
- The full diff is available via the
cli-v3.0.67...cli-v3.0.68comparison on GitHub.
What this means
The teams.db fix matters more than the catalog changes. Re-saving whole state on each streamed chunk is a write-amplification problem that scales with session length and team size, which is exactly the profile of long-running multi-agent coding work. Cline's multi-agent features are now being used in sessions long enough to expose this kind of storage cost, and a database that reached gigabytes suggests some users were hitting it in practice.
The default-model changes show how quickly provider defaults move. Two of the five changes point to the same model, GPT-6.1 Sol, on different hosts (DigitalOcean and Ofox), and the Nvidia default is a Flash-tier DeepSeek variant. That suggests the CLI is tracking the models providers are actively serving rather than pinning older, stable choices. Teams that depend on reproducible behavior should pin an explicit model instead of relying on provider defaults, since those defaults can change between patch releases.
Related Articles
NVIDIA launches Nemotron for Science with Nemotron 3.5 Super VL and a chemistry dataset of nearly 50,000 questions
NVIDIA introduced Nemotron for Science on November 16, 2026. It pairs the open Nemotron 3.5 Super VL model with scientific evaluations, post-training recipes and LitmusBench v0.2, a chemistry dataset of nearly 50,000 questions.
Google Adds Shareable Interactive Sites to Gemini Notebook, Rolling Out Globally on Web
Google announced on November 16, 2026 that Gemini Notebook can turn a notebook's sources into interactive sites. Outputs now include custom visual explainers, dashboards and project trackers, and the feature is rolling out globally on the web.
Perplexity launches Monitor API: web monitoring priced at $1 per 1,000 successful runs
Perplexity launched the Monitor API on November 16, 2026. It gives developers scheduled topic searches and webpage-change tracking, with filtered updates delivered through the API or signed webhooks. Web monitoring costs $1 per 1,000 successful runs.
Siri AI can now answer Health data questions in iOS 27.3 beta 2
Apple's iOS 27.3 beta 2, released to developers on November 16, lets Siri AI answer questions about Health app data. The assistant had been blocked from that data since Siri AI launched in iOS 27.0, while the older Siri Classic could access it.
Comments
Loading...