Sourcebot growth playbook

Make the wall a guided decision instead of a surprise. (1) Add a pricing FAQ and a short /upgrade page: "Upgrading from v5 with more than five users?" with the exact behaviour, the 14-day Pro trial, the price for their user count and how existing subscriptions keep their terms. (2) On the restricted-mode screen, put "Start 14-day Pro trial" as the first button, with the seat count already filled in, and a second option to pick five users. (3) Show the same notice in the app before the upgrade (a banner in v5.x releases) and in the release notes. (4) Track how many restricted deployments start a trial, convert or roll back.

The team has already proved value: those deployments have more than five people using them. A clear price, a no-card trial and an honest explanation turn a lock-out into a purchase decision while the pain of losing search is fresh, and they reduce the chance that upset admins pin old releases or go to a rival.

Q: I'm upgrading from v5 and have more than five users. What happens?
A: v6 needs an activation code. The free Basic plan covers up to five active users. If more than five people use your deployment, it opens in restricted mode: sign-in and Settings work, but search, Ask, the API and MCP pause until you either start a free 14-day Pro trial (no card needed) or reduce to five active users. Nobody is removed automatically. Pro is $20 per user a month billed yearly. Existing subscriptions keep their terms until renewal.
[Start 14-day Pro trial]  [See the upgrade guide]  [Talk to us about Enterprise]

• Admins feel locked out and complain publicly

  Lead with the facts and the free options (five users, the trial), and offer a short grace extension on request.

• Trials start but do not convert

  Book a 15-minute check-in on day 7 of every trial above 20 seats.

• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source
• Your docsFree trial: deploy Sourcebot first, then start the Pro trial from onboarding or Settings, License, by confirming an email; no credit card; one trial per deployment; the deployment must be able to send its Service Ping.Source
• Your websitev6 changelog (October 8, 2026): Ask Sourcebot and MCP, previously paid, are now free for teams of up to five users; existing subscriptions keep their current terms until the end of the current term.Source

Publish /compare/sourcegraph: a fair, dated table of deployment (single container vs enterprise install), pricing and minimums, free tier, code hosts, search, natural-language questions, MCP for agents, permission syncing, SSO/SCIM and support. Add a cost example for 10, 50 and 200 engineers, a short "when Sourcegraph is the better fit" section, and a migration offer. Link it from the pricing page and the README.

Teams searching for a Sourcegraph alternative have budget and intent, and price is the clearest difference. An honest page also gives AI assistants a clean source when asked for alternatives.

• Claims go out of date

  Date the page, link sources and invite corrections.

• Readers see bias

  Say plainly where Sourcegraph is stronger, such as certifications and batch changes.

• Competitor researchSourcegraph: "The context layer for your entire codebase" with Code Search, Deep Search, batch changes and an MCP server; "200+ enterprise engineering teams"; cites benchmark results such as 38% faster coding agent execution; one Enterprise plan starting at $16K with a minimum annual contract via contact sales, self-hosted or single-tenant cloud, no free tier listed; lists SOC2 Type II and ISO27001.See the research Source
• Hacker NewsShow HN "Sourcebot, an open-source Sourcegraph alternative" (October 1, 2024): 259 points and 59 comments; commenters asked whether it would go closed source like Sourcegraph, and asked for Gitea/Forgejo, self-hosted Bitbucket and Perforce support.Source
• Your website"Sourcebot vs OpenGrok" (May 14, 2025) is the only comparison page; it still marks Sourcebot's blame and history support "coming soon", while the homepage now lists git history and blame, and it predates the v6 plans.Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source

Either remove the link today, or make /articles the home for evergreen buyer content: the Sourcegraph and OpenGrok comparisons, the GitLab search post, and the guides in ideas 9 and 12. Add it to the sitemap once it has at least five entries.

Small trust signals matter for a tool that asks for access to source code, and a sitewide link is a strong internal signal for whatever it points to.

• A section that stays thin

  Only relaunch it with five or more entries.

• Your websiteSitemap (32 URLs, checked November 10, 2026): homepage, /code-context, /pricing, terms, privacy, 20 changelog entries and 7 blog URLs; no case study, use-case, integration, comparison or code-host pages. The footer on every page links to /articles, which is not in the sitemap and lists no articles.Source
• Your websiteBlog: six listed posts: November 2024 (searching across GitLab projects), May 2025 (an AI code review agent; Sourcebot vs OpenGrok), February 2026 (bug zero with Linear) and October 2026 (the closed-source post; one-shotting an Unreal Engine plugin). Neither the blog nor the changelog offers an email or RSS subscription; the changelog's only follow option is X.Source

Add a short notice at the top of the README: what is public (earlier releases and their license), what v6 is, where images are published, the free five-user plan and a link to /upgrade. Update the GitHub description, the YC profile and the job post to "self-hosted" wording. Pin an issue that answers the common questions.

Consistent facts lower friction in procurement and calm the long-time community the Show HN thread came from. It also points the repo's steady traffic at the new trial instead of an old license.

• Restating the change revives criticism

  Keep it factual and lead with what stays free.

• #1 Turn the v6 five-user wall into a trial path: tell teams above five users what happens before they upgrade, and offer Pro in the same screen

• GitHubRepo sourcebot-dev/sourcebot: 4.0k stars, 380 forks, 1,525 commits, marked Public; the sidebar links an "MIT license" and the "fair-source" topic; the README says "a self-hosted code context layer for humans and AI agents" and links to pricing, but does not mention v6, the move to a private repository or the five-user Basic plan.Source
• Y CombinatorYC profile (F25, team size 3, San Francisco): still describes "an open source code understanding platform" used by thousands of engineers at companies including NVIDIA, Red Hat, Wikimedia and Arista Networks, and says it can be deployed in minutes because it is open source and on-prem; the founders previously worked at Microsoft, EA, Ubisoft, Google and Meta.Source
• HiringChief of Staff (San Francisco, $120K to $200K, 0.10% to 1.00% equity, the only open role): "Prospect key accounts", "Drive enterprise pilots to adoption and conversion", run operations and launch cross-functional growth initiatives, with exposure to enterprise sales, customer success, hiring and marketing; wants someone "AI pilled" who automates their work. The post still calls Sourcebot open source.Source
• Your websiteBlog, "Sourcebot is Now Closed Source" (October 8, 2026): "Starting with v6, future Sourcebot development will take place in a private repository"; "previous releases will remain publicly available"; "We're no longer confident that open-core can sustain the business we want to build"; the company wants to build ways for agents to capture and retrieve knowledge "without making that work immediately available to competitors", and expects companies to run "millions of agents".Source

Update the table with every shipped feature, add a row for agent context over MCP, a row for pricing (both free for small teams; Sourcebot Pro for permissions and analytics), an "updated" date and a short migration section. Link to the Sourcegraph page and the trial.

A stale "coming soon" on a comparison tells the evaluator the page, and maybe the product, is not maintained. Fixing it is an hour's work on a page with existing search traffic.

• Unfair claims about OpenGrok

  Check against OpenGrok's current docs.

• Your website"Sourcebot vs OpenGrok" (May 14, 2025) is the only comparison page; it still marks Sourcebot's blame and history support "coming soon", while the homepage now lists git history and blame, and it predates the v6 plans.Source
• Your websiteHomepage: "code understanding for humans and agents"; a self-hosted tool that indexes GitHub, GitLab, Bitbucket, Azure DevOps, Gerrit and Gitea and runs as a single Docker container; calls to action are npx setup-sourcebot, the public demo, the deployment guide and Book a demo; testimonials from a Red Hat security engineer ("we had the job done in an afternoon instead of a week"), Oegema Logistics+ and AF Automations.Source

Add a "Get release notes by email" field to the changelog, blog, docs and the in-app What's new panel. Send one email per release with breaking changes first, and a monthly digest. Offer the same in an RSS feed.

Admins who know what is coming upgrade with less friction and fewer support tickets, and the list becomes the company's own channel for launches, trials and surveys.

• Too many emails

  Offer releases-only or monthly digest.

• Your websiteBlog: six listed posts: November 2024 (searching across GitLab projects), May 2025 (an AI code review agent; Sourcebot vs OpenGrok), February 2026 (bug zero with Linear) and October 2026 (the closed-source post; one-shotting an Unreal Engine plugin). Neither the blog nor the changelog offers an email or RSS subscription; the changelog's only follow option is X.Source
• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source

Build a simple weekly report from existing ping data: deployments at 4-5 active users, deployments in restricted mode, fast-growing weekly actives and large repository counts. Where the admin has confirmed an email (trial or activation), send a helpful, personal note: what Pro adds, a seat quote and an offer to help with permission syncing. Keep it to data the docs already say is collected, and say in the privacy page how it is used.

Usage-led selling is the most efficient motion for a three-person team: it focuses scarce sales time on teams that have already chosen the product.

• Outreach feels like surveillance

  Only contact admins who gave an email, use counts not names, and explain in the privacy page.

• #1 Turn the v6 five-user wall into a trial path: tell teams above five users what happens before they upgrade, and offer Pro in the same screen

• Your docsService Ping: every deployment sends, every 24 hours, an install ID, hostname, version, deployment type, total, daily, weekly and monthly active user counts, indexed repository count and a system snapshot; "No sensitive data (source code, AI inputs/outputs, credentials, user information, etc) is ever transmitted."Source
• HiringChief of Staff (San Francisco, $120K to $200K, 0.10% to 1.00% equity, the only open role): "Prospect key accounts", "Drive enterprise pilots to adoption and conversion", run operations and launch cross-functional growth initiatives, with exposure to enterprise sales, customer success, hiring and marketing; wants someone "AI pilled" who automates their work. The post still calls Sourcebot open source.Source
• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source

Write three stories: the Red Hat security task, an AI-agent story (the logistics customer that calls Sourcebot fundamental to its AI work), and one large-codebase deployment from the logo list if they agree. Each has the codebase size, code hosts, time to deploy, what changed and a quote. Put them on /customers, in the pilot playbook and on the pricing page.

Security, IT and platform leaders buy on peer proof, and named logos without stories leave the evaluator to guess the value.

• Large companies refuse to be named

  Offer anonymised stories with industry and size.

• Your websiteHomepage: "code understanding for humans and agents"; a self-hosted tool that indexes GitHub, GitLab, Bitbucket, Azure DevOps, Gerrit and Gitea and runs as a single Docker container; calls to action are npx setup-sourcebot, the public demo, the deployment guide and Book a demo; testimonials from a Red Hat security engineer ("we had the job done in an afternoon instead of a week"), Oegema Logistics+ and AF Automations.Source
• Your website/code-context: "Multi-repo code context for AI agents"; "Deployed in" NVIDIA, Shutterstock, SeatGeek, Chip, Arista and Red Hat; backed by Y Combinator, Liquid 2 and Pioneer Fund; three steps: deploy and index, connect a coding agent over MCP, then the agent can list branches, grep, resolve symbols, read files and get diffs across all repositories; code, indexes, queries and AI inputs and outputs stay in the customer's environment.Source
• Your websiteSitemap (32 URLs, checked November 10, 2026): homepage, /code-context, /pricing, terms, privacy, 20 changelog entries and 7 blog URLs; no case study, use-case, integration, comparison or code-host pages. The footer on every page links to /articles, which is not in the sitemap and lists no articles.Source

Write a migration guide per tool: mapping of concepts (code hosts, repo lists, search contexts, auth), a side-by-side search syntax table and a small converter that turns an existing repo list or config into Sourcebot's config file. Link each guide from the matching comparison page.

Switching cost is the main reason teams stay on an older tool. A guide that makes migration a one-afternoon task removes it.

• Converter breaks on edge cases

  Start with the common config formats and show a manual path.

• #2 Publish a Sourcebot vs Sourcegraph page: free for five and $20 a seat against an Enterprise plan from $16K
• #5 Refresh the only comparison page: OpenGrok still shows blame as "coming soon"

• Your docsDocs index: deployment, sizing, code hosts (including Gerrit, Gitea, generic Git hosts and local repos), MCP, Agent Skills, a review agent, connectors and four version-upgrade guides; no guide for moving from another code search tool and no case studies; the roadmap is GitHub issue #459 and support is via GitHub issues.Source
• Hacker NewsShow HN "Sourcebot, an open-source Sourcegraph alternative" (October 1, 2024): 259 points and 59 comments; commenters asked whether it would go closed source like Sourcegraph, and asked for Gitea/Forgejo, self-hosted Bitbucket and Perforce support.Source
• Competitor researchSourcegraph: "The context layer for your entire codebase" with Code Search, Deep Search, batch changes and an MCP server; "200+ enterprise engineering teams"; cites benchmark results such as 38% faster coding agent execution; one Enterprise plan starting at $16K with a minimum annual contract via contact sales, self-hosted or single-tenant cloud, no free tier listed; lists SOC2 Type II and ISO27001.See the research Source

Pick 30 realistic multi-repo tasks on public codebases (find all callers of an API across repos, fix a bug that spans two services, answer an architecture question). Run the same coding agents with only local repos, then with Sourcebot over MCP. Publish success rate, time, tokens and the full harness so anyone can rerun it. Repeat with each release.

Platform and AI leads buying agent infrastructure want evidence, and an open harness is more credible than a vendor's single number.

• Results are modest

  Publish anyway and show which task types gain most.

• Competitor researchSourcegraph: "The context layer for your entire codebase" with Code Search, Deep Search, batch changes and an MCP server; "200+ enterprise engineering teams"; cites benchmark results such as 38% faster coding agent execution; one Enterprise plan starting at $16K with a minimum annual contract via contact sales, self-hosted or single-tenant cloud, no free tier listed; lists SOC2 Type II and ISO27001.See the research Source
• Your website/code-context: "Multi-repo code context for AI agents"; "Deployed in" NVIDIA, Shutterstock, SeatGeek, Chip, Arista and Red Hat; backed by Y Combinator, Liquid 2 and Pioneer Fund; three steps: deploy and index, connect a coding agent over MCP, then the agent can list branches, grep, resolve symbols, read files and get diffs across all repositories; code, indexes, queries and AI inputs and outputs stay in the customer's environment.Source
• Your websiteBlog: six listed posts: November 2024 (searching across GitLab projects), May 2025 (an AI code review agent; Sourcebot vs OpenGrok), February 2026 (bug zero with Linear) and October 2026 (the closed-source post; one-shotting an Unreal Engine plugin). Neither the blog nor the changelog offers an email or RSS subscription; the changelog's only follow option is X.Source

Publish /security with a data-flow diagram, a table of every outbound connection (ping, license, optional model provider) and what each sends, how to run fully offline with the Enterprise license, and the security features by plan (SSO, permission syncing, SCIM, audit logs). Offer a downloadable questionnaire pack and state the company's compliance plans honestly.

Security review is often the longest step in buying a self-hosted tool. Answering the standard questions up front shortens pilots and reassures admins worried about the telemetry change.

• Over-promising compliance

  State only what is in place today and the dates of anything planned.

• Your docsService Ping: every deployment sends, every 24 hours, an install ID, hostname, version, deployment type, total, daily, weekly and monthly active user counts, indexed repository count and a system snapshot; "No sensitive data (source code, AI inputs/outputs, credentials, user information, etc) is ever transmitted."Source
• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source
• Competitor researchSourcegraph: "The context layer for your entire codebase" with Code Search, Deep Search, batch changes and an MCP server; "200+ enterprise engineering teams"; cites benchmark results such as 38% faster coding agent execution; one Enterprise plan starting at $16K with a minimum annual contract via contact sales, self-hosted or single-tenant cloud, no free tier listed; lists SOC2 Type II and ISO27001.See the research Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source

Create one page per self-managed host: the search gap on that platform, a two-minute setup snippet from the existing docs, screenshots on a public repo hosted there, and a trial link. Start with Gerrit and Gitea, where competition is thinnest, then Bitbucket Data Center and Azure DevOps Server.

Niche, high-intent searches such as "Gerrit code search across repositories" have few good answers, and the people searching run exactly the self-hosted infrastructure Sourcebot is built for.

• Low search volume

  Use deployment counts by host to decide where to stop.

• Your docsDocs index: deployment, sizing, code hosts (including Gerrit, Gitea, generic Git hosts and local repos), MCP, Agent Skills, a review agent, connectors and four version-upgrade guides; no guide for moving from another code search tool and no case studies; the roadmap is GitHub issue #459 and support is via GitHub issues.Source
• Hacker NewsShow HN "Sourcebot, an open-source Sourcegraph alternative" (October 1, 2024): 259 points and 59 comments; commenters asked whether it would go closed source like Sourcegraph, and asked for Gitea/Forgejo, self-hosted Bitbucket and Perforce support.Source
• Your websiteSitemap (32 URLs, checked November 10, 2026): homepage, /code-context, /pricing, terms, privacy, 20 changelog entries and 7 blog URLs; no case study, use-case, integration, comparison or code-host pages. The footer on every page links to /articles, which is not in the sitemap and lists no articles.Source
• Your websiteBlog: six listed posts: November 2024 (searching across GitLab projects), May 2025 (an AI code review agent; Sourcebot vs OpenGrok), February 2026 (bug zero with Linear) and October 2026 (the closed-source post; one-shotting an Unreal Engine plugin). Neither the blog nor the changelog offers an email or RSS subscription; the changelog's only follow option is X.Source

Write one short page per supported agent: what the agent cannot see today, the exact config snippet, three example prompts that use cross-repo search, and a GIF-free screenshot. Submit Sourcebot's MCP server to the main public MCP directories and registries, pointing each listing at the matching page.

It meets developers at the moment they are configuring their agent, which is when they feel the local-only limit most.

• Config formats change often

  Date each snippet and check monthly.

• Your website/code-context: "Multi-repo code context for AI agents"; "Deployed in" NVIDIA, Shutterstock, SeatGeek, Chip, Arista and Red Hat; backed by Y Combinator, Liquid 2 and Pioneer Fund; three steps: deploy and index, connect a coding agent over MCP, then the agent can list branches, grep, resolve symbols, read files and get diffs across all repositories; code, indexes, queries and AI inputs and outputs stay in the customer's environment.Source
• Your docsDocs index: deployment, sizing, code hosts (including Gerrit, Gitea, generic Git hosts and local repos), MCP, Agent Skills, a review agent, connectors and four version-upgrade guides; no guide for moving from another code search tool and no case studies; the roadmap is GitHub issue #459 and support is via GitHub issues.Source

Write a standard pilot: day 1 kickoff with agreed success measures (searches per engineer, Ask questions answered, agent calls, time saved on one named task), week 1 deploy and index with the sizing guide, week 2 permission syncing and SSO, week 3 a mid-point report from Pro analytics, week 4 a business review and an order form. Package it as a one-page PDF and a shared checklist.

Pilots that start with agreed measures convert far more often than open-ended trials, and the same data feeds the customer stories in idea 8.

• Customers will not commit to measures

  Offer default measures they can accept as-is.

• HiringChief of Staff (San Francisco, $120K to $200K, 0.10% to 1.00% equity, the only open role): "Prospect key accounts", "Drive enterprise pilots to adoption and conversion", run operations and launch cross-functional growth initiatives, with exposure to enterprise sales, customer success, hiring and marketing; wants someone "AI pilled" who automates their work. The post still calls Sourcebot open source.Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source
• Y CombinatorYC profile (F25, team size 3, San Francisco): still describes "an open source code understanding platform" used by thousands of engineers at companies including NVIDIA, Red Hat, Wikimedia and Arista Networks, and says it can be deployed in minutes because it is open source and on-prem; the founders previously worked at Microsoft, EA, Ubisoft, Google and Meta.Source

Generate a one-page monthly report inside each deployment from its own analytics: active users, searches, Ask questions, MCP calls, top repositories, and month-on-month trend. Send it to admins by email (inside the customer's environment, using their configured email), with a "forward to your manager" summary at the top.

It makes value visible to the budget holder, supports seat expansion and reduces churn at renewal, without sending any customer data to Sourcebot.

• Low usage looks bad

  Include tips to raise usage, and flag low use to the Chief of Staff early.

• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source
• Your docsDocs index: deployment, sizing, code hosts (including Gerrit, Gitea, generic Git hosts and local repos), MCP, Agent Skills, a review agent, connectors and four version-upgrade guides; no guide for moving from another code search tool and no case studies; the roadmap is GitHub issue #459 and support is via GitHub issues.Source

List every tag in Tag Manager with an owner and a decision. Keep analytics and the one or two ad networks that actually drive trials; remove the rest; load the remaining pixels after interaction. Fix the long cache lifetimes and 274 KB of unused JavaScript flagged by PageSpeed.

Fewer tags mean a faster first impression for technical buyers, cleaner attribution and fewer third-party cookies on a site that sells privacy and self-hosting.

• Losing ad attribution

  Keep the pixels for networks with active campaigns.

• PageSpeedPageSpeed (November 10, 2026): real visitors pass Core Web Vitals (field LCP about 1.4 s, CLS 0). Lab mobile score 69 (desktop 67): Total Blocking Time 977 ms on mobile and 1,799 ms on desktop, 878 KB of JavaScript of which 274 KB is unused; 12 third-party services weigh 519 KB, with Google Tag Manager alone 341 KB and 553 ms of blocking; ad and conversion pixels from LinkedIn, Reddit, X and Google load on the homepage.Source

Add five clickable example searches and two example questions that show cross-repo power, a persistent bar with "Run this on your own repos: npx setup-sourcebot" and "Free for up to 5 users", and a short MCP example. Keep Book a call for larger teams. Tag each link so demo-to-deploy can be measured through the install ID.

The demo already attracts the most hands-on visitors. Giving them a self-serve path converts curiosity into installs without adding sales load.

• Fewer demo calls booked

  Keep Book a call visible for teams above 20 engineers.

• Your websitePublic demo: opens without sign-in as "Guest"; no example queries; the only call to action is "Book a call"; no deploy, pricing or trial link.Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source

Turn the sign-in wall on shared links into an invitation: show the question and a preview of the answer, "Your colleague shared this from Sourcebot", and a one-click request to join (routed to the admin, with the seat price shown to admins on Pro). Add a "share to Slack" button on answers and a "skill of the week" prompt for admins.

Internal sharing is the natural way seat-based developer tools spread, and it costs nothing to acquire a colleague who already has a reason to look.

• Previews expose code to the wrong people

  Show only the question to signed-out users and respect permission syncing.

• Your websiteAgent Skills (v5.1.0, July 10, 2026): reusable team instructions for Ask Sourcebot, importable from markdown or from skill files in indexed repositories, and shareable to a workspace where admins manage the catalog.Source
• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source
• Your docsDocs index: deployment, sizing, code hosts (including Gerrit, Gitea, generic Git hosts and local repos), MCP, Agent Skills, a review agent, connectors and four version-upgrade guides; no guide for moving from another code search tool and no case studies; the roadmap is GitHub issue #459 and support is via GitHub issues.Source

Pilot a managed, single-tenant Sourcebot per customer in the region they choose, with the same features, at a premium per seat (for example Pro plus a platform fee). Start with ten design partners on cloud code hosts, measure ops cost per tenant, and publish the security model alongside idea 11.

It removes the biggest activation step for teams without platform engineers and raises revenue per seat, while keeping the self-hosted promise for those who need it.

• Ops load overwhelms a small team

  Cap the pilot at ten tenants and automate provisioning first.

• Undercuts the self-hosted message

  Position it as single-tenant and keep self-hosted as the default.

• #11 A security review pack for self-hosted buyers: what leaves the deployment, and what never does

• Your docsFree trial: deploy Sourcebot first, then start the Pro trial from onboarding or Settings, License, by confirming an email; no credit card; one trial per deployment; the deployment must be able to send its Service Ping.Source
• Your websitePricing page (checked November 10, 2026): Basic is free for up to 5 users with code search, SSO, code navigation, Ask Sourcebot, REST API/MCP and community support; Pro is $20 per user a month billed yearly ($25 monthly) with unlimited users, permission syncing, org management and analytics, and a "Start 14-day free Pro trial" button that opens the docs; Enterprise is custom (SCIM, audit logs, Slack channel, support engineer and SLAs, custom MSA, offline license) via Book a call. The FAQ covers trying, self-hosting, code storage and code hosts, but not the activation code or what happens to a team above five users. No managed cloud plan is offered.Source
• Competitor researchSourcegraph: "The context layer for your entire codebase" with Code Search, Deep Search, batch changes and an MCP server; "200+ enterprise engineering teams"; cites benchmark results such as 38% faster coding agent execution; one Enterprise plan starting at $16K with a minimum annual contract via contact sales, self-hosted or single-tenant cloud, no free tier listed; lists SOC2 Type II and ISO27001.See the research Source

Introduce an agent add-on: service accounts for agents and CI pipelines, priced by monthly MCP and API calls in simple tiers, on top of human seats. Show live agent usage in analytics. Test with three Pro customers who already hit high MCP volumes, then publish it on the pricing page.

It aligns price with the value the company is betting on, opens budget from platform and AI teams rather than per-developer tooling, and gives the 10,000-request limit a clear upgrade path instead of a wall.

• Usage pricing scares buyers

  Use fixed tiers with soft limits rather than metered surprises.

• #15 Email Pro admins a monthly usage report they can forward to their manager

• Your websiteBlog, "Sourcebot is Now Closed Source" (October 8, 2026): "Starting with v6, future Sourcebot development will take place in a private repository"; "previous releases will remain publicly available"; "We're no longer confident that open-core can sustain the business we want to build"; the company wants to build ways for agents to capture and retrieve knowledge "without making that work immediately available to competitors", and expects companies to run "millions of agents".Source
• Your docsV5 to V6 upgrade guide: the free plan becomes Basic, which "requires an activation code"; without a valid license the deployment enters restricted mode, blocking search, Ask, file browsing, the REST API and MCP; a deployment with more than five active users "stays in restricted mode" until it is at five or fewer or upgrades; no users are removed automatically; anonymous access is removed; API and MCP requests are limited to 10,000 per user per day; the telemetry opt-out variable is removed.Source
• Competitor researchGreptile: "The AI Code Reviewer"; free for one developer, then $1 per credit with no seat fees; self-hostable with Docker Compose or Kubernetes.See the research Source

Index the 500 most-used open-source repositories and expose them on a public site with crawlable pages per repository ("Search the <project> codebase") and a free, rate-limited public MCP endpoint for agents. Every page carries "Do this on your private repos: free for 5 users".

Each repository page can rank for its own long-tail searches, and developers who use it daily learn the product before they need it at work.

• Hosting cost grows faster than conversions

  Start with 50 repositories and expand only on proven deploys.

• #17 Make the public demo end in a deploy: example queries and a self-host button

• Your websitePublic demo: opens without sign-in as "Guest"; no example queries; the only call to action is "Book a call"; no deploy, pricing or trial link.Source
• Hacker NewsShow HN "Sourcebot, an open-source Sourcegraph alternative" (October 1, 2024): 259 points and 59 comments; commenters asked whether it would go closed source like Sourcegraph, and asked for Gitea/Forgejo, self-hosted Bitbucket and Perforce support.Source
• Your websiteSitemap (32 URLs, checked November 10, 2026): homepage, /code-context, /pricing, terms, privacy, 20 changelog entries and 7 blog URLs; no case study, use-case, integration, comparison or code-host pages. The footer on every page links to /articles, which is not in the sitemap and lists no articles.Source

Interview three studios about how they search engine and game code today. If the pain is real, build a read-only Perforce connector for one design partner, publish a joint story, and a page for game studios that links the Unreal post.

Supporting a code host rivals ignore can give Sourcebot an entire vertical, and the founders' industry background makes the first conversations easier.

• Distraction from the core market

  Time-box to one quarter and one partner.

• #14 A 30-day enterprise pilot playbook with success criteria agreed on day one

• Hacker NewsShow HN "Sourcebot, an open-source Sourcegraph alternative" (October 1, 2024): 259 points and 59 comments; commenters asked whether it would go closed source like Sourcegraph, and asked for Gitea/Forgejo, self-hosted Bitbucket and Perforce support.Source
• Your websiteBlog: six listed posts: November 2024 (searching across GitLab projects), May 2025 (an AI code review agent; Sourcebot vs OpenGrok), February 2026 (bug zero with Linear) and October 2026 (the closed-source post; one-shotting an Unreal Engine plugin). Neither the blog nor the changelog offers an email or RSS subscription; the changelog's only follow option is X.Source
• Y CombinatorYC profile (F25, team size 3, San Francisco): still describes "an open source code understanding platform" used by thousands of engineers at companies including NVIDIA, Red Hat, Wikimedia and Arista Networks, and says it can be deployed in minutes because it is open source and on-prem; the founders previously worked at Microsoft, EA, Ubisoft, Google and Meta.Source
• Your websiteHomepage: "code understanding for humans and agents"; a self-hosted tool that indexes GitHub, GitLab, Bitbucket, Azure DevOps, Gerrit and Gitea and runs as a single Docker container; calls to action are npx setup-sourcebot, the public demo, the deployment guide and Book a demo; testimonials from a Red Hat security engineer ("we had the job done in an afternoon instead of a week"), Oegema Logistics+ and AF Automations.Source