a context-mode Platform product
Engineering productivity from AI tool usage.
Teams running the OSS plugin — now upgrading to Insights
The problem
You signed the seat license. The plugin is rolled out. Engineering is moving faster — or it looks like it is. Then the EM Slack pings: "Ahmet has been stuck on auth.ts for three days." The CTO wants a number for the next board deck. The CISO wants to know what the AI actually touched.
There is no observability today. The agent sessions happen on each laptop, in each editor, behind each API key. You see the bill, not the work. You can guess which teams over-rely on AI. You cannot prove it.
Burning roughly $60K / year across 50 engineers on stuck context cycles — error loops, rewrites, re-prompts — that no dashboard surfaces.
The product
The context-mode OSS plugin already runs on every developer machine. Every Claude Code, Cursor, Copilot, Codex, Gemini session emits structured events through its hook system. Insights ingests those events server-side and runs 177 patterns over them. The findings surface in three places — the same data, three doors in.
https://platform.context-mode.com/mcp — eight tools, role-tuned schemas, OAuth.Nothing on the developer laptop changes. The hook fires fire-and-forget on every tool call. POST /api/v1/events. The data lives in D1, queried by the platform Worker. Same plugin. Same workflow. Three new doors.
Walkthrough
The same data layer answers radically different questions depending on who is asking. Insights is role-aware — the schema for every tool is dynamically narrowed per token.
"What is my AI tool investment actually returning? Show me org-level metrics for the last 30 days."
engagement_health(scope: "org", period: "30d")
Org-wide ROI snapshot: 17/17 active. Adoption 100%. Estimated savings 34 hours / week / engineer (high confidence, based on Think-in-Code substitution rate). Top inactive: 0. compare_teams(period:"30d") shows backend leading frontend by +18% productivity, -12% error rate.
→ Renews the seat license. Board deck has a number.
"Bu hafta takımlarımda kim engellenmiş? Pazartesi 1:1'leri hazırlıyorum."
find_blockers(scope: "team", period: "7d", limit: 5)
Top finding: "Ahmet K. — error_cycle_member_high in auth.ts, 12 retries, 0 commits, severity high. Suggested: pair-program review." Two more: a frontend dev hit rejection_loop on a CSS refactor, a backend dev is on blocked_on > 14d for a migration. Berke's managed_teams column scopes the result to Backend, Frontend, Design only — server filters, schema enforces.
→ 1:1 schedule rewritten. Pattern broken before sprint end.
"How am I doing this week? Am I better or worse than the team average?"
how_am_i(period: "7d")
Personal snapshot: productivity 78, quality 71. Strengths: parallel_tool_use_high, think_in_code_adopter. Growth area: read_before_edit_skipped — 6 times this week. Peer benchmark (anonymised): you are at the 62nd percentile on the team. Recommendation: "When editing a file you have not opened this session, read it first — your error_cycle rate drops 40% in the sessions where you do."
→ Adjusts workflow. No one else sees the score.
Same data. Three scope ceilings. Selin sees the org. Berke sees only Backend, Frontend, Design. Ahmet only sees himself. Role is not a UI filter — it is a schema constraint. Member tokens literally cannot pass scope: "org". The enum lacks the value.
Tool catalog
Each tool is a first-class registration — none wraps another. Descriptions follow the WHEN / WHEN NOT / SCOPE / RETURNS / EXAMPLE rubric so the agent picks the right one without re-prompting.
Coverage
Patterns are the unit of insight. Each one is a SQL query plus a severity rule plus a suggested action. Built by analysing real session telemetry — not invented.
Every pattern carries an evidence_count on the wire — when Insights says "Ahmet is stuck", you can drill to the 12 event rows that backed the finding.
Pricing
Engineering organisations have predictable headcount but unpredictable AI usage — one big PR can burn 100 tool calls in five seconds. Pricing reflects that.
Audit Product and Risk Score Product on the roadmap — built for Cem (CISO) and Hakan (DevOps Lead) respectively. Free design-partner slots open.
Security & access
Most analytics products bolt RBAC on top with middleware checks. Insights enforces it one layer earlier — at the MCP tool schema itself. A Member's find_blockers tool literally does not have "org" in its scope enum. The check is structural, not policy.
managed_teams column lists which teams they see — 1 team = Team Lead, N teams = EM.Documented in ADR-0007. Server narrows queries by managed_teams for Manager, by member_id for Member. Owner skips the filter.
Install
If you are already on the OSS plugin, you have already done step one. Step two is one dashboard click.
npx context-mode init
curl -fsSL https://platform.context-mode.com/install.sh | bash -s -- $TOKEN
The bootstrap token is one-hour-scoped and generated by the dashboard. Script detects which AI assistants you have and adds the platform MCP server to each — Claude Code via claude mcp add, Cursor via deeplink, Codex via config. Restart the editor. First MCP call triggers OAuth in the browser. Token in keychain. Done.
FAQ
Structured session events emitted by the OSS plugin's hook system: tool calls (name, duration, exit status), prompt boundaries, file paths touched, error / retry counts, and git operations. No prompt content, no file content, no source code. The plugin operates locally; only the structured event envelope ships.
Owner sees the org. Manager sees the teams listed in their managed_teams column. Member sees themselves plus anonymised peer benchmarks. Enforced at the MCP schema layer per ADR-0007 — a Member token literally cannot pass scope: "org".
Yes. The OSS plugin is fully independent — local FTS5 knowledge base, no cloud calls, no telemetry. Insights is opt-in: you generate a token in the platform dashboard and the same plugin starts forwarding events. Remove the token, forwarding stops.
Enterprise tier. Cloudflare Workers run anywhere with the right account; the D1 schema and the pattern catalog are open. Self-hosted SOC2 / FedRAMP work is custom — book a call.
Two add-on Products on the roadmap. Audit turns the event stream into compliance reports for the CISO persona (Cem). Risk Score scores AI-authored changes inside a PR for the DevOps Lead persona (Hakan), exposed as a GitHub Action. Neither ships in v1 — design partners welcome.
Ready to look?
If you already run the OSS plugin across your team, we can have your first three findings on the call. Bring an EM. We will show find_blockers against your real session data.
Onboarding 5 design-partner orgs
Free Team-tier access during the design phase. Shape the patterns we ship next.