Here’s a simple test for evaluating “self-hosted” products: can you use their AI features without going through their cloud? With Sourcegraph, the answer is no (see the screenshot from their docs). You can deploy Sourcegraph inside your own infrastructure, but if you want to use Deep Search your code and prompts are sent to their model proxy. This creates a weird situation where you've purchased a tool because you can self-host and not share your data, but then to use one of its main features you have to share your data. Sourcebot is Bring Your Own Key, allowing you to connect to any LLM provider you like. Your data stays within your environment and you're not charged an extra margin on usage. We've worked with several teams that have made the switch, if your team is hitting this issue visit our website to book a call with me directly.
About us
Sourcebot is a open source code understanding platform for massive codebases. We’re 10x-ing developer and AI agent productivity by building the tools to wrangle code complexity at scale.
- Website
-
https://www.sourcebot.dev/
External link for Sourcebot
- Industry
- Software Development
- Company size
- 2-10 employees
- Type
- Privately Held
Employees at Sourcebot
View 4 employees at Sourcebot
or
Updates
-
We just launched Sourcebot V5, our biggest release yet. Ask Sourcebot can now fetch context from external tools such as Jira, Linear, Notion, and more. Combine Sourcebot's deep multi-repo code analysis with context from all the other tools your team is using. This change is our first step in making Sourcebot a place where engineering teams ask questions not only on their code, but across their entire organization. We've also refreshed our design with a persistent sidebar, as well as added full git blame & history support. Try it now by running 𝗻𝗽𝘅 𝘀𝗲𝘁𝘂𝗽-𝘀𝗼𝘂𝗿𝗰𝗲𝗯𝗼𝘁
-
What we shipped this week (Mar 14 - Mar 21): - Admins can now restrict API key creation and usage to organization owners only. - MCP search_code now supports filtering by file path using regular expressions. - Redis TLS is now supported. - Also: agent max step count raised from 20 to 100, MCP read-only annotations for Cursor Ask mode compatibility, ask_codebase hidden when no LLM providers are configured, and SMTP individual env var support.
-
-
A developer at one of our customers built a version tracker for their Spring Boot services. The old version hit the Bitbucket API, made recursive calls across every repo, and took 10 minutes to run. They couldn't run it during business hours without triggering rate limits. It was a manual job they ran when they needed it. The new version queries Sourcebot's search API and takes 30 seconds. It's wired to a cron job that runs every night and pushes results to Splunk automatically. The performance improvement is real, but what it revealed was more interesting: building on top of a powerful code search API made automation obvious. The 10-minute version existed because the right data wasn't accessible in a clean way. Once it was, everything downstream got easier. The value of an indexed codebase isn't just in the search, the index becomes a data source for everything else you want to build on top of it.
-
What we shipped this week at Sourcebot (Mar 7 - Mar 14): - [EE] Multi-owner support. Promote and demote admins directly from your org settings. - REST API documentation. OpenAPI documentation is now available for our public search, repo, and file browsing API. - [EE] JumpCloud is now supported as an SSO identity provider. - Also: MCP authentication improvements, Bitbucket permission sync reliability fixes, etc.
-
We heard from users that our MCP was difficult to configure, here's how we fixed it: Sourcebot MCP gives agents context across a companies entire codebase, so it's critical that setting this up is as low friction as possible. First, we moved our MCP server from a separate npm package into the application itself. Not only does this prevent the developer from pulling a separate npm package, it's also one less package we need to maintain. We also added OAuth support to the MCP client, allowing users to easily auth directly in the web app instead of creating an API key. Here's a quick demo showing the new MCP configuration flow in Cursor:
-
Sourcebot reposted this
I used to spend hours answering security questionnaires for Sourcebot. Enterprise DevSecOps teams would send a giant list of niche questions around how we manage data, connect to external systems, handle authentication, etc. I'd have to investigate each question and write a detailed response and answer any follow ups they have. It only recently dawned on me that I could use Sourcebot to automate this entire process. Now, I just feed every question they ask into Ask Sourcebot and have it write a detailed response that includes citations directly into our codebase. The best part is I can just share the link to the response directly, and their team can ask follow up questions directly in the same thread! If your codebase is public on GitHub you can prepend "ask" to your repo url to ask questions in Sourcebot directly for free. If not, you can deploy Sourcebot on-prem and connect to any code host platform (links in the comments).
-
Sourcebot reposted this
Traditional code search is great when you know what you're looking for - but what if you don't? We just shipped an AI search assistant which helps you generate traditional code search queries using natural language. For when you want the convenience of asking in natural language, with the determinism and speed of traditional code search. Try it out today! Link to docs below
-
Sourcebot reposted this
We just shipped Bitbucket permission syncing. Enterprises using Bitbucket can now ensure their repo visibility rules are reflected in Sourcebot. Sourcebot now supports syncing permissions with GitHub, GitLab, Bitbucket Cloud, and Bitbucket Data Center. Enterprise teams that use multiple code host platforms can ensure that repo access controls across all of them are respected when users search for code in Sourcebot's UI or through our MCP server. Link to the release in the comments!
-
-
Sourcebot reposted this
PRs on Sourcebot have 5x'd in the past two months. If you're not seeing similar improvements you're likely not taking advantage of recent agent developments. Cursor announced today that 35% of their PRs are written by agents. These agents have shifted from being assistants to now being one of the main ways modern teams ship code. Two of the most important aspects to utilize agents like this are through proper orchestration and context. Running a single agent in an IDE is dead. Modern teams are managing teams of agents, both locally and on the cloud, to tackle issues simultaneously. A simple setup to do this locally is to use multiple Claude Code sessions using git worktrees. For cloud setups, we're currently using Linear connected to Cursor cloud agents. These agents won't be able to get the job done without the proper context. You should be connecting your agents with all of the sources your engineers have (jira, slack, etc) to give it the right context. To enable these agents to understand your codebase in multi-repo setups, you can connect them to Sourcebot using our MCP server. Reviewing and validating these changes is still a huge problem, and is something we're investing time into to ensure our agentic product velocity isn't capped moving forward.
-