Sourcebot – Self-hosted Perplexity for Your Codebase

TL;DR

TL;DR: Sourcebot is a self-hosted, open-source code intelligence tool that lets you ask natural-language questions about your codebase, search across all repos, and get answers backed by inline citations.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: sourcebot.dev ← verified
• Source repository: github.com/sourcebot-dev/sourcebot ← README read end-to-end
• Live demo: app.sourcebot.dev ← verified public deployment
• Docs: docs.sourcebot.dev ← deployment docs read
• License: NOT YET SET — repository has no LICENSE file. GitHub API returns NOASSERTION. Do not assume a license; check with the maintainers before using in proprietary projects.
• HN launch thread: not listed in README; searched via HN Algolia (no confirmed launch thread found)

What Is Sourcebot?

Sourcebot is a self-hosted tool that helps developers and AI agents understand a codebase. It indexes your repositories and exposes two main interfaces:

1. Ask Sourcebot — Ask natural-language questions and get answers with inline citations that link directly to the relevant source code.
2. Code Search — Search across all repos and branches using regex, filters, and boolean logic. Source-level navigation (goto definition, find references) works across all repositories.

The project positions itself as a self-hosted alternative to hosted code intelligence services. It is built with Docker Compose in mind and requires PostgreSQL and Redis as backing services.

Setup Workflow

Prerequisites

• Docker and Docker Compose
• At least 4 GB RAM recommended
• A config file (config.json) for code host connections (GitHub, GitLab, etc.)
• Environment variables for secrets (AUTH_SECRET, SOURCEBOT_ENCRYPTION_KEY, DATABASE_URL, REDIS_URL)

Step 1: Download the Docker Compose file

curl -o docker-compose.yml https://raw.githubusercontent.com/sourcebot-dev/sourcebot/main/docker-compose.yml

Step 2: Configure code host access

Edit the config.json file to connect your code hosts. The full configuration reference is at docs.sourcebot.dev/docs/configuration/code-host.

A minimal GitHub config looks like this:

{
  "codeHosts": [
    {
      "type": "github",
      "id": "my-org",
      "token": "ghp_your_github_token"
    }
  ]
}

Step 3: Set environment variables

Generate secrets using OpenSSL, then add them to a .env file:

AUTH_SECRET=$(openssl rand -base64 33)
SOURCEBOT_ENCRYPTION_KEY=$(openssl rand -base64 24)

Then in your .env file:

AUTH_SECRET=your_generated_auth_secret
SOURCEBOT_ENCRYPTION_KEY=your_generated_encryption_key
DATABASE_URL=postgresql://postgres:password@postgres:5432/postgres
REDIS_URL=redis://redis:6379

Step 4: Start the containers

docker compose up -d

Sourcebot starts on port 3000 by default. The initial setup will index the configured repositories, which may take several minutes depending on codebase size.

Deeper Analysis

Architecture

Sourcebot uses a web interface backed by a search indexing layer. It maintains a PostgreSQL database for query storage and Redis for caching. Repositories are cloned and indexed locally — no data leaves your infrastructure.

What sets it apart

• Self-hosted — no external API calls, all code and data stays on your servers
• Multi-repo support — indexes across all connected repos simultaneously, surfacing results from different repositories in a single answer
• Inline citations — answers reference exact file locations and line numbers
• IDE-level navigation — goto definition and find references work across repo boundaries

Limitations

• No built-in authentication provider; you configure your own AUTH_SECRET
• Initial indexing can be slow for large monorepos
• No official cloud-hosted option — fully self-hosted only

Practical Evaluation Checklist

• [ ] Docker Compose starts cleanly with docker compose up -d
• [ ] Web UI accessible on port 3000 after startup
• [ ] GitHub token works and repos appear in the dashboard
• [ ] Ask Sourcebot returns answers with clickable inline citations
• [ ] Code search returns results with regex and language filters
• [ ] No sensitive data in logs or error messages
• [ ] Indexing progress visible during initial setup

Security Notes

• All secrets must be generated and stored securely — never commit .env files
• GitHub tokens used for repo access should be scoped to read-only where possible
• Sourcebot does not currently support SSO or fine-grained access control
• Database and Redis should be on a private network or bound to localhost

FAQ

Q: Does Sourcebot work with GitLab or Bitbucket?

A: Yes. The configuration supports multiple code host types. See the code host configuration docs for the full list.

Q: How long does initial indexing take?

A: It depends on repository size and number of repos. Small repos index in minutes; large monorepos can take an hour or more. Indexing runs in the background after startup.

Q: Can I use Sourcebot without an internet connection?

A: Yes. Once the Docker images are downloaded and repos are cloned locally, Sourcebot operates entirely offline.

Q: Is there a way to restrict access to certain repositories?

A: Access control is configured at the code host level through token scoping. There is no per-repo permission layer within Sourcebot itself at this time.

Conclusion

Sourcebot fills a specific niche: teams that want Perplexity-style code intelligence but need it self-hosted. The inline citations and cross-repo search are its strongest features, and the Docker Compose setup makes it accessible to teams without a dedicated platform engineering function. If you have been evaluating Sourcegraph Cloud or GitHub Copilot for code understanding and need the data to stay in-house, Sourcebot is worth a weekend evaluation.

Try the public demo at app.sourcebot.dev before committing to a self-hosted deployment.