Khoj
Khoj gave its assistant long-term memory, then started winding down its own cloud.
A side-by-side editorial comparison of OpenRouter and Roo Code — release velocity, themes, recent moves, and the top alternatives to consider.
OpenRouter starts building above the router, not just wider
Recent output splits into two kinds of posts: endpoint documentation for modalities beyond chat, with image generation at /api/v1/images and transcription at /api/v1/audio/transcriptions, and explainers on routing economics such as prompt caching with sticky routing and the per-provider price spread on DeepSeek. The one genuine product addition is Classifiers, which tags every generation in a workspace against a customer-defined taxonomy. The rest is positioning around one base URL, one key, one bill.
Roo Code's visible releases are CLI point versions aimed at driving the agent from a script.
The four available entries are all Roo Code CLI releases from early March 2026, versions 0.1.14 through 0.1.17, shipped over roughly 36 hours. Every one of them touches programmatic control: a flag to supply a custom UUID session ID when creating tasks, a flag to choose which shell runs inline commands, and two fixes to stdin-stream mode covering follow-up message routing after a clarification question and truncated command output before the done event.
Recent output splits into two kinds of posts: endpoint documentation for modalities beyond chat, with image generation at /api/v1/images and transcription at /api/v1/audio/transcriptions, and explainers on routing economics such as prompt caching with sticky routing and the per-provider price spread on DeepSeek. The one genuine product addition is Classifiers, which tags every generation in a workspace against a customer-defined taxonomy. The rest is positioning around one base URL, one key, one bill.
The modality additions follow the aggregator playbook: same key, same routing controls, one more content type. Classifiers is a different move, building an OpenRouter-owned layer above the passthrough where the value is attribution rather than access. The caching, sticky-routing, and price-spread posts all point the same direction, shifting the pitch from reaching every model to controlling what agents spend.
Expect further workspace-level governance built on the same taxonomy Classifiers introduces, most likely budgets or policies attached to logs and Activity analytics. More modality endpoints are likely too, but on this evidence they will be routine additions rather than direction changes.
The four available entries are all Roo Code CLI releases from early March 2026, versions 0.1.14 through 0.1.17, shipped over roughly 36 hours. Every one of them touches programmatic control: a flag to supply a custom UUID session ID when creating tasks, a flag to choose which shell runs inline commands, and two fixes to stdin-stream mode covering follow-up message routing after a clarification question and truncated command output before the done event.
This is the surface of the product being hardened for automation rather than for interactive use. Session IDs that a caller can set and validate, deterministic stdin streaming, and a selectable shell are the pieces you need to embed an agent in someone else's pipeline. The releases are small, fast and narrow, and the extension itself is not represented in this feed — so this reads as one component's release channel, not the product's.
The run of stdin-stream fixes suggests further work on that path, but the newest entry here is from March 2026 and nothing has followed, so the current state of the CLI is not visible from this feed.
Other ai-assistants products tracked by Sparkpulse, ranked by recent ship velocity. Each card links to a full editorial trajectory and lets you pivot into a head-to-head comparison with either OpenRouter or Roo Code.
Khoj gave its assistant long-term memory, then started winding down its own cloud.
Cline made agentic compaction the default and put its own models on a free tier.
Copilot's agent surfaces are being pulled under the same enterprise policy plane as the editor
DocsBot is moving its agent out of the website widget and into Slack, with redaction guarding the way in.
Sourcegraph is rebuilding code search as the retrieval layer coding agents rent.
LangGraph's 1.2 line is maintenance work on state durability, not new ground.
See all OpenRouter alternatives → · See all Roo Code alternatives →
Latest ship moves from both products, interleaved chronologically. ⚡ = editorial spark.
They serve adjacent needs but don't currently overlap on shipped themes. OpenRouter is currently shipping more aggressively (velocity 6.3 vs 0.0), with 1 editorial sparks in the last 30 days against 0. See the at-a-glance table above for a side-by-side breakdown of velocity, recent sparks, and editorial themes.
Sparkpulse doesn't pick a winner — we score release velocity, not feature parity. OpenRouter is currently shipping more aggressively (velocity 6.3 vs 0.0), with 1 editorial sparks in the last 30 days against 0. For your specific use case, the alternatives sections above list other ai-assistants products to evaluate alongside.
Top OpenRouter alternatives in ai-assistants are ranked by recent ship velocity. Browse the "OpenRouter alternatives" section above for the current picks, or visit /alternatives/openrouter for the full list with editorial commentary on each.
Top Roo Code alternatives in ai-assistants are ranked by recent ship velocity. Browse the "Roo Code alternatives" section above for the current picks, or visit /alternatives/roo-code for the full list with editorial commentary on each.