Sourcegraph vs Sourcebot

Where Sourcebot wins

There is a plan for you, and it costs nothing

The single most useful fact about Sourcebot is that its free tier is not a trial. Self-hosted Basic has no per-user fee and no seat count, and it includes the parts people actually use every day: search across every repository and branch, regex and symbol queries, git history and blame, and a file explorer. Compare that with Sourcegraph in 2026, where the only published plan starts at $16K and the free option was discontinued on 23 July 2025. For a fifteen-person team that lost Cody Pro, this is the difference between an evaluation and a procurement cycle.

It is the most actively developed project in the category

Sourcebot shipped v5.1.5 on 31 July 2026 and had commits landing on 6 August 2026, the day this comparison was compiled. Among self-hosted code search projects that is unusual: OpenGrok releases a few times a year, Hound has not tagged a release since May 2023, and livegrep has never tagged one. If you are betting a piece of developer infrastructure on a small project, release cadence is the risk indicator that matters most, and this one currently reads well.

You can read the code you depend on

FSL is not OSI open source, and this page says so twice for a reason. But the whole source is published, you can build it, patch it and audit it, and the restriction it imposes is narrow: you may not sell it as a competing product. For the overwhelming majority of internal deployments that restriction never binds, while the ability to read the code does. That is precisely the position teams lost when Sourcegraph made its core repository private in August 2024.

Deployment is a config file, not a project

A Sourcebot install is a Docker Compose file plus a JSON configuration listing your connections and repositories. There is no separate database to operate, no search cluster to size, and no indexer schedule to design. Sourcegraph's self-hosted deployments were, by contrast, a Kubernetes conversation. Treat vendor claims about setup time as vendor claims, but the shape of the deployment is verifiable from the repository and it is genuinely small.

Where Sourcegraph still wins

Batch Changes has no equivalent, and it is not a small feature. Being able to write one change, apply it across four hundred repositories, and then track the resulting pull requests to completion in a single view is the reason many platform teams bought Sourcegraph in the first place. Code Insights is the same story in a different direction: turning a search query into a tracked metric that leadership watches over quarters. Neither is replicable with a search UI, and any migration plan that quietly omits them is not a plan.

Precise navigation is the subtler gap. Sourcegraph builds SCIP indexes from real compilation, so jump-to-definition and find-references are exact across repository boundaries. Sourcebot's navigation, which sits behind the Pro tier, derives from the search index instead. That is good enough to read unfamiliar code comfortably and not good enough to answer "prove nothing else calls this" during a security review. And there is a case where Sourcegraph's pricing genuinely wins: at a thousand engineers, $16K is roughly sixteen dollars a head, which is less than Sourcebot Pro at $20 per user per month for a single month. The enterprise price is punishing at fifteen people and unremarkable at a thousand.

Migrating Sourcegraph to Sourcebot

The path most teams take:

1. Inventory first. List your saved searches, code monitors, Batch Changes and Code Insights dashboards, and mark which ones have no equivalent. Search moves for free; those four do not move at all. Deciding consciously to drop a code monitor is fine, discovering its absence in an incident is not.
2. Point Sourcebot at the same hosts. The swap itself is a connections file listing GitHub, GitLab, Bitbucket, Gitea or Gerrit organisations and the branches to index, then disk. There is no export step, because a code search index is derived data, and this is why this migration is cheaper than its reputation.
3. Run both for one cycle. Keep the Sourcegraph contract through a full quarter and let people search in both. You are looking for the query shapes that get worse, usually language-specific symbol searches and anything that relied on precise references, and you can only find them from real use.
4. What does not move. Every Sourcegraph permalink pasted into an issue, pull request or design doc over the last several years will 404 the day the instance goes away. Nobody counts these until they are gone. Export the ones inside your own wiki and rewrite them before the cutover, and accept that the ones inside old pull request threads are lost.