Sold by: Takara
Deployed on AWS
Free Trial
Takara's ds1-code, powering Miru, is a CPU-based code embedding and retrieval model that gives AI coding agents precise, semantic access to a codebase without burning tokens on wasteful search. Delivering over 50% token reduction, 40%+ lower token costs, and up to 60% faster retrieval, Miru works underneath Claude Code, Cursor, Codex, GitHub Copilot, and other coding agents to cut inference spend while preserving generated code quality.
Overview
Takara's ds1-code is a CPU-based embedding model purpose-built for understanding code at the semantic level, underpinning Miru, a code-retrieval infrastructure layer for AI coding agents. Rather than dumping entire code bases into an LLM's context window and hoping it identifies what's relevant, Miru combines ds1-code with hybrid retrieval - BM25, lexical search, vector search, and ranking - to deliver the exact function, module, or dependency chain an agent needs in a single retrieval step. Because ds1-code runs on CPU rather than GPU, it avoids GPU procurement, queuing, reservation planning, and regional rollout complexity, while scaling close to developers globally. The result is over 50% token reduction and 40%+ lower token costs across coding agent workflows, with retrieval tasks up to 60% faster and less turns required than conventional approaches, without any impacting the quality of generated code. Miru can be tried via a public API to benchmark value against real code generation workflows, and deployed as a SageMaker endpoint in a customer's own AWS environment - keeping source code and software IP inside a controlled environment.
NOTE: ds1-code is designed to be exclusively used with Miru (https://github.com/takara-ai/miru-code )
Highlights
- Cut coding agent token spend. Modern AI coding agents burn most of their token budget not on writing code, but on finding it - reading files, re-loading previously seen context. Miru, powered by ds1-code, replaces that brute-force search with precise semantic retrieval, delivering exactly the code an agent needs in a single step, cutting token consumption across real coding workflows. It works underneath the agents teams already use, including Claude Code, Cursor, Codex, and GitHub Copilot.
- CPU-based architecture that scales without GPU bottlenecks. ds1-code is built to run entirely on CPU, eliminating the need for GPU reservation that typically constrain AI infrastructure rollouts. Letting Miru scale without the lead times or cost overhead of GPU-based retrieval, while still delivering search that is up to 60% faster than conventional approaches. This means a lower-cost path to giving every coding agent fast, high-quality retrieval, wherever developers are located.
- Sovereign deployment for code and IP that must stay under your control. Source code is one of an organisation's most sensitive assets. Try Miru instantly via the public API to benchmark real workflows, then deploy as a SageMaker endpoint in your AWS environment. Code and IP stay inside your controlled environment at every stage, giving CIOs and security teams a sovereign deployment model that meets data residency needs without giving up the speed or cost advantages of ds1-code.
Details
Sold by
Categories
Delivery method
Latest version
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Free trial
Try this product free for 21 days according to the free trial terms set by the vendor.
Pricing is based on actual usage, with charges varying according to how much you consume. Subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost/host/hour |
|---|---|---|
ml.c7i.large Inference (Batch) Recommended | Model inference on the ml.c7i.large instance type, batch mode | $0.64 |
ml.c7i.large Inference (Real-Time) Recommended | Model inference on the ml.c7i.large instance type, real-time mode | $0.64 |
ml.c5.xlarge Inference (Batch) | Model inference on the ml.c5.xlarge instance type, batch mode | $1.20 |
ml.c5.2xlarge Inference (Batch) | Model inference on the ml.c5.2xlarge instance type, batch mode | $2.46 |
ml.c5.4xlarge Inference (Batch) | Model inference on the ml.c5.4xlarge instance type, batch mode | $4.92 |
ml.c5.9xlarge Inference (Batch) | Model inference on the ml.c5.9xlarge instance type, batch mode | $11.04 |
ml.c5.18xlarge Inference (Batch) | Model inference on the ml.c5.18xlarge instance type, batch mode | $22.02 |
ml.t2.medium Inference (Real-Time) | Model inference on the ml.t2.medium instance type, real-time mode | $0.42 |
ml.t2.large Inference (Real-Time) | Model inference on the ml.t2.large instance type, real-time mode | $0.84 |
ml.c5.large Inference (Real-Time) | Model inference on the ml.c5.large instance type, real-time mode | $0.90 |
Dimensions summary
You pay by the hour for the compute instance that runs the DS1 code embedding model. Charges depend on two choices: the instance type and the inference mode. Batch mode runs on compute-optimized c5, c6i, and c7i instances. Real-time mode adds smaller t2 instances plus the c5, c6i, and c7i families. Within each family, sizes range from large through xlarge and higher multiples. Larger sizes carry more capacity and cost more per hour. This model runs on CPU instances, so no GPU option applies. Billing scales with the hours each chosen instance stays active.
Top-of-mind questions for buyers
One HostHrs unit is one hour that a chosen instance stays active running the DS1 code embedding model. You pay per running hour, per instance. If you run several instances at once, each accrues its own hourly charge on the same invoice.
Both meter running instance-hours. Real-time mode hosts a live endpoint that stays active for on-demand requests, so hours accrue while it runs. Batch mode processes grouped jobs on compute-optimized instances. Real-time mode also offers smaller t2 instances that batch mode does not.
Charges apply per hour while the instance is active. A real-time endpoint that stays running accrues hours even without traffic. Stopped instances do not accrue software charges, but underlying AWS storage or resource fees may still apply separately.
takara.ai
Helpful?
Vendor refund policy
Refunds are furnished in line with the EULA only. Please contact support@takara.ai for assistance.
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Amazon SageMaker model
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Version release notes
This version improves performance by skipping infer queue for static embedding models, and additionally now supports newer family EC2 instances for deployment.
Additional details
Inputs
- Summary
This model is for use with Miru only.
- Limitations for input type
- This model should be accessed and used in conjunction with Miru available for download here: https://github.com/takara-ai/miru-code. Using this model standalone is not supported.
- Input MIME type
- application/json
application/json
application/json
Support
Vendor support
Please email support@takara.ai for customer support for next day response.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Takara's DS1 Embedding Model is a high-speed, low-computing text embedding solution, leveraging static embeddings for superior performance. Offering exceptional speed and near-OpenAI accuracy, it's ideal for applications demanding swift semantic responses, such as speech-to-speech, betting, and gaming. With an API compatible with OpenAI, DS1 ensures a seamless upgrade experience.
Takara's DS1 Embedding Model is a high-speed, low-computing text embedding solution that now supports multiple languages, leveraging static embeddings for superior performance. Offering exceptional speed and near-OpenAI accuracy across global markets, it's ideal for applications demanding swift semantic responses, such as speech-to-speech, betting, and gaming. With an API compatible with OpenAI, DS1 ensures a seamless upgrade experience without language barriers.
Text embedding model optimized for general-purpose (including multilingual) retrieval/search and AI applications. 32K context length.
Rich multimodal embedding model that can vectorize interleaved text, content-rich images, and video. 32K context length.
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.