Unblocked

Mentioned, not specified.

Yes on certification: Unblocked states it is 'AICPA SOC 2 Type II' certified (alongside CASA Tier II and GDPR compliance), and says the full SOC 2 report is available at trust.getunblocked.com. On source-code storage, the answer is only partial: the code-review FAQ says 'Unblocked integrates with your existing tools through secure OAuth connections. We respect all your existing permissions and never store your code permanently,' and the ai-info page says Unblocked 'indexes your repositories and connected data sources and keeps them in sync.' However, the site never explicitly specifies what is persisted — whether only embeddings and metadata, or transient/derived copies of source code during indexing. So a buyer can conclude code is not stored permanently, but the exact storage model (embeddings + metadata vs. copies) is not documented.

Mentioned, not specified.

Yes on posting directly, yes on per-repo configuration, but only per-user (not per-team) off control is documented. 1) Posts comments directly on PRs — YES. The docs state: "Once enabled, Unblocked automatically reviews each open pull request when it's created or updated. The AI analyzes code changes for potential bugs and issues, then posts inline code comments directly in your PR when issues are found." If no actionable issues are found, it "will leave an approved review on the PR." Marketing copy confirms: "Unblocked Code Review posts automated, context-aware review comments on every pull request." 2) Restrict to specific repos — YES, at the configuration level. The FAQ states: "You can configure review rules, exclude specific file patterns, and adjust the level of feedback to match your team's needs. Per-repository configuration is supported." Risk-based auto-approval likewise uses "configurable per-repo risk policies," set via a .unblocked/risk-policies.yaml file added to a repository. Analytics docs also describe per-repo rollout ("Teams typically start with one or two repos and expand from there"). However, the exact admin steps for enabling/disabling Code Review on a given repo are not in the retrieved pages. 3) Turn it off per team — NOT DOCUMENTED. Only per-user control exists in the retrieved docs: each developer can toggle "Receive automated code reviews as comments on my pull requests" under Settings > Personal Settings > Preferences, and the docs explicitly note the limit: "When disabled, you won't receive automated comments, but team-level reviews still run for other team members." No team-level or admin-level on/off switch for Code Review appears anywhere in the retrieved documentation, so a buyer cannot confirm from the site whether Code Review can be disabled for an entire team.

Mentioned, not specified.

Unblocked's site documents no explicit cap on how many repositories a workspace can connect, and no per-repo size limit — but it also never states 'no limit' outright, so a hard ceiling can't be ruled out from the site alone. The closest evidence: Unblocked's context-engine blog says 'the largest organizations on Unblocked run well over 100,000 repositories with tens of thousands of engineers,' implying no practical cap at least into six figures. For per-repo size, the GitHub and GitHub Enterprise docs say only that 'if your repositories have a lot of code, documentation, and discussions, then data ingestion may take longer to process' — i.e., large repos are handled with slower ingestion rather than being rejected or excluded. The repo-selection flows (GitHub, Bitbucket, GitLab) describe choosing which repos to ingest — repos active in the last 6 months are preselected, forks and archived repos are excluded — with no stated maximum. Notably, the site asserts 'no limit to how many repositories you can index' only about GitHub Copilot Chat in its comparison post, not about Unblocked itself.

Mentioned, not specified.

Two-part answer. (1) Beta status — not explicitly resolved: the only beta reference on the site is the announcement post 'Introducing Unblocked MCP Server (Beta)' (getunblocked.com/blog/mcp-beta/), and no retrieved page states the server has reached GA. However, the current integration pages present it as a live, standard feature — a single remote server at https://getunblocked.com/api/mcpsse with OAuth — with no beta caveat, so it appears to be out of beta, but the site never says so explicitly. (2) Clients — clearly answered: 'Any MCP-compatible client (Devin, Codex, VS Code, Claude Desktop, and others) can connect to the remote server with OAuth' at https://getunblocked.com/api/mcpsse. The integration pages specifically name Claude Code, Cursor, GitHub Copilot, and Codex ('To give Claude Code, Cursor, GitHub Copilot, Codex, or another MCP-compatible AI coding agent access...'), and the beta announcement also lists Windsurf. For clients without native support, the docs describe Settings → Add Custom MCP, adding the server, and authenticating via OAuth.

Mentioned, not specified.

It's a hybrid, and the site gives one concrete number — but only on the GitLab docs page. There, Unblocked states that merge requests and their discussions are "synchronized in real time, so they are always up to date," while source code runs on a schedule: "Unblocked indexes source code from your repository's default branch at least once every 8 hours. Answers and search — including through Unblocked MCP — use this index, so changes on other branches appear after they merge to the default branch." So after merging a PR, discussion content is reflected immediately (the changelog adds that new PR comments and reviews are "re-indexed within seconds of being posted, instead of waiting for the periodic sync"), but code-level answers reflect the merged code only within the next index cycle — up to ~8 hours per the GitLab docs. The marketing pages only describe this loosely: the ai-info page says Unblocked "keeps them in sync as your team pushes changes" with the main branch as "the source of truth," and the GitHub integration blurb says repos are "indexed, kept in sync on every push" — neither gives a latency figure, and the 8-hour cadence is stated only for GitLab, so the post-merge latency for GitHub (or other sources) is not pinned down.

Mentioned, not specified.

The site publishes concrete, quantified token-savings numbers and describes how they were measured — but none of the figures in these chunks are attributed to customers. They come from Unblocked's own controlled tests, Unblocked's own production traffic, or third parties: 1. Unblocked's own controlled test (same prompt, same model, different context): a 42% token cut at equal output, with the context-grounded run finishing in 174 seconds vs 237 seconds and a higher first-pass merge rate — measured by running identical prompts with and without Unblocked's context. 2. Unblocked's own production data (Opus → GLM migration): rate-card math implied ~95% savings (2.35B tokens served for ~5% of the equivalent Opus cost, caching accounted for on both sides), but running the same code review on the same commit, same PR, same prompt with each model showed the real-world saving was 68%, because GLM burned more tokens per task (average tokens per call rose from ~18,500 to ~24,700 over the migration window). Method stated: run identical workloads end to end and compare invoices, not rate cards, backed by a fleet-wide ledger. 3. Measurement methodology Unblocked uses/advocates: cost measured per task (not per model call) via a ledger recording provider, token counts, duration, time-to-first-token and task ID; and 'token yield' — useful output per token spent, tied to merged PRs or resolved tickets (cost per merged PR) rather than raw token consumption. 4. Third-party numbers the site cites (not customers): Anthropic prompt caching up to 90% cost reduction on stable prompt prefixes; RouteLLM ~85% cost reduction on MT-Bench at ~95% of GPT-4 quality; GitHub MCP tool allow-listing cutting schema overhead ~92% (~4,200 vs ~55,000 tokens, Ken Imoto's third-party measurement). So: concrete reductions and measurement methods exist in abundance, but no chunk provides a customer-named or case-study measurement of token spend reduction — the closest to 'real customers' is Unblocked running itself in production.

Mentioned, not specified.

Yes on export/stream, with one gap. (1) UI export: Audit Logs are Enterprise-plan only, and you can 'export your team's full log history by selecting the ellipse button in the top right corner, and clicking Export as CSV' from Settings → Team Settings → Audit Logs. (2) SIEM streaming: the Public API has a 'Security Events' feed explicitly described as 'Retrieve a feed of team events for downstream ingestion into a security information and event management (SIEM) system. The feed currently includes audit logs. Poll the feed to collect new events' — implemented as GET /audit-logs with `since`/`after` parameters and stable keyset pagination ('Cursors use stable keyset pagination and never expire'). Auth behavior: with Personal Access Tokens, 'Team admins can export all audit logs for their team; non-admin members receive 403 Forbidden'; team-wide API keys return all audit logs for the team. Note that event recording only started September 3, 2025, so historical backfill is limited. Admin events captured: team-level changes like renames or deletions; data source connections, disconnections, and modifications; member adds, removals, and role changes; modifications to security features (including Data Shield and SSO); license assignments, trial extensions, and billing updates; CI provider configuration changes and API token updates — plus a detailed per-data-source event table (e.g., Asana workspaces added/deleted and project selection changes, Confluence site/space changes, Bitbucket Data Center enterprise application creation). Query events: not documented — the docs scope tracked events to 'your team's settings or data source management,' and neither the docs nor the launch blog lists user queries/questions as audited events, so whether query activity is captured or flows to your SIEM is unstated.

Mentioned, not specified.

Partially. For canceling/closing the account, Unblocked's privacy policy is specific: "Customer data is retained for as long as the account is in active status. Data enters an 'expired' state when the account is voluntarily closed. Expired account data will be retained for 30 days. After this period, the account and all related data will be removed." For involuntarily suspended accounts, there is a 30-day grace period; after 60 days the suspended account is closed, its data enters the "expired" state, and it is "permanently removed 30 days thereafter (except when required by law to retain)." Note this policy speaks at the account level ("all related data") and does not break out indexed code vs. chat history. For the other scenario — disconnecting an individual repository while keeping the account — the site gives no deletion timeline: the data-source docs explain how to add or remove connected repositories (e.g., selecting/unselecting repos under Settings > Data Sources > GitHub Enterprise) but say nothing about when the indexed data for a removed repo is purged. The closest adjacent statement is in the API reference, "Deleting a collection will also remove all the associated documents," but that covers API collections, not repository disconnection.

Mentioned, not specified.

Yes on the features: Unblocked supports SAML SSO with any SAML 2.0-compliant identity provider — the SSO overview explicitly lists Okta, Google Workspace, Microsoft Entra ID, AWS Identity Center, Ping Identity, and 'other SAML 2.0 compliant providers' — and it supports SCIM for automatic user and group provisioning and deprovisioning ('Unblocked is a fully compliant SCIM 2.0 service provider and supports user and group provisioning from any SCIM-compliant identity provider'). Setup is documented per IdP (generate a SCIM API token under Team Settings → Security → SCIM User and Group Provisioning, then configure your IdP), and Unblocked also exposes a full SCIM 2.0 API. On plan gating: no page in the retrieved site content states that SAML SSO or SCIM is restricted to a specific plan tier. Marketing pages list 'SAML SSO with SCIM provisioning' as part of Unblocked's security posture (alongside SOC 2 Type II, CASA Tier II, RBAC, audit logs) and explicitly gate only on-premises / in-your-cloud deployment to the Enterprise tier — SSO and SCIM themselves are never tied to a tier. So feature availability is confirmed, but the site does not publish an explicit plan-tier requirement for SSO/SCIM.

Mentioned, not specified.

Support channels: yes, and they differ by plan. The Platform plan ($29/user/month billed annually) includes "Web-based support" per the pricing page. The Enterprise plan includes "Everything in Platform, plus: ... Dedicated customer success, Priority support via Slack and Zoom" — i.e., Enterprise buyers get priority support delivered through Slack and Zoom channels plus a dedicated customer success contact. The plan-comparison table's Support section also lists "Web Support, Premium Support, Dedicated Account Manager, Dedicated Slack Channel, Custom Terms, Guided Security Review, Invoiced Billing" as support-related features, though the flattened table does not make clear which plan each row maps to. Response-time SLA: no. Nowhere in the retrieved site content — pricing page, docs, or blog — is any response-time commitment, SLA percentage, or guaranteed response window mentioned for the Enterprise plan or any other plan. The site names the Enterprise support channels (Slack, Zoom, dedicated success) but never quantifies how fast responses will be.

Mentioned, not specified.

Unblocked does not publicly name the specific LLMs behind its answers. Its product pages say only that it 'uses connections to the systems used to build your application and optimized LLMs to help you understand the nuances and specifics of your codebase,' with a disclaimer that 'Unblocked utilizes LLMs to generate responses, which may result in occasional inaccurate or nonsensical responses.' An engineering blog post ('Routing LLM traffic across inference providers by cost, speed and reliability', Sep 29, 2026) adds that Unblocked routes traffic across multiple inference providers selling 'the same open-weight models,' using an adaptive router that 'keeps us on the cheapest provider that is fast and healthy, and moves our traffic when that changes, without an engineer involved' — but no specific model names are given anywhere. On the second part of your question: nothing on the site indicates that customers can choose a model or route requests through their own provider/key. The routing described is Unblocked's internal, automatic infrastructure ('without an engineer involved'), not a customer-facing option, and no BYO-model or BYO-provider feature is documented. Based on the site's content, customer-selectable models or bring-your-own-provider routing appear unsupported, though the site never states this explicitly.