One context.
Every agent.
Your team's context, structured once. Every coding agent gets the rules file it expects — in the right format, at the right path, always current.
Specs drift.
Rules rot.
Agents ship the gap.
Context scatters across tools. Each agent reads only its own rules file — and only some of the time.
"The space between what a team knows and what an agent knows is where every project fails."
- Wiki page× Never reaches the agent
- Chat thread× Lost after 14d
- One-off prompt× Vanishes on refresh
- Shared doc× Read by humans, not agents
From repo to every agent.
Your team in the loop.
Connect once. AI flags the gaps, your team fills them in, the CLI ships the result.
Link your repo
The CLI reads your repo locally and builds a structured context. Your source never leaves your machine.
AI surfaces the gaps
ContextHub reviews what it can't infer from your repo and queues targeted questions — sharp prompts, not fuzzy chat.
Anyone answers, anywhere
Questions queue on the web and in the terminal. Answer in seconds — every decision is recorded and folded into the shared context.
Ship with ch sync
One command pushes the latest graph and pulls every agent file — in the format it expects, at the path it expects, never stale.
Two letters.
Every agent.
ch keeps your agents in sync. Push your latest graph, pull agent files, answer questions, and add agents — all from the command line.
Write context
as a team.
Planners, leads, and anyone with answers — draft, comment, and refine. AI reads alongside and flags anything it needs clarified.
More than a rules file in git.
Six things that make context actually work across your whole team.
Structured, not scattered
Context is organised around the actual parts of your product — extracted from your repo and kept in sync.
AI that asks the right questions
AI reviews your context and only asks where something is genuinely unclear — sharp prompts, not noise.
Efficient by design
Context is compacted before it reaches your agent — fewer tokens, lower cost, same quality. Still human-readable in your repo.
Agents are a set, not a switch
ch agent add codex. Claude Code and Codex render side by side — the whole team mixes agents without file churn.
Restorable snapshots
Every save snapshots the graph. Restore any version. No branches, no merge conflicts.
Repo stays on your machine
Graph builds locally. Only the graph uploads — never your source, never your secrets.
We're building ContextHub with five design-partner teams. A few slots are open.
Early access · direct line to the founder · pricing locked at launch rates
- seed → Series A
- 2+ coding agents in use
- ships weekly
- any stack
One context. One sync.
Every agent.
Today's agent. Next month's. The one nobody has shipped yet. Every one gets the file it expects.
- 01 context built from your repo, reviewed by AI, kept in sync
- 01 workspace your team writes once, restores anytime
- 01 sync graph up, agent files down, every path
Your code never leaves your machine.
-
Client-side analysis
Your repo is read locally. Only structure uploads — never source, never secrets.
-
Encrypted, never trained on
TLS in transit, encrypted at rest. Your data never trains anything.
-
Roles + audit trail
Owner / Editor / Viewer. Every answer and approval logged, append-only, on Team.
-
You own the context
One command exports everything — the context and every decision. No lock-in by design.
Free for solo. Pay when the team joins.
Planned pricing — locked at these rates for design partners. Hosted AI on every plan; your code never uploads.
Free
- 3 contexts, solo workspace
- Hosted AI with monthly limits
- Every agent format
- 30-day version history
- One-command export
Pro
- Unlimited contexts
- Hosted AI, higher limits
- Full version history + restore
- Plans that compile to agent task lists
- Drift check for CI
Team
- Everything in Pro
- Invites + roles (Owner / Editor / Viewer)
- Append-only audit trail
- Plan approval workflow
- Priority support + onboarding
Common Questions
Why not just commit a rules file to git?
You can. Then you do it again for the next agent. Then you forget which one's stale. The graph is the single source — every agent's file falls out of it.
Which agents are supported?
Claude Code, Cursor, GitHub Copilot, Codex (via AGENTS.md — which also covers Zed, Amp, Aider, and friends), and Gemini CLI. Adding a new one is a one-file config change, not a release.
Does my code leave my machine?
No. Analysis runs locally. Only the structural map — names, paths, relationships — and the markdown you write upload. Source bodies and secrets never do, and we publish exactly what we store.
How is this different from a wiki?
Wikis are for humans and go stale silently. ContextHub knows what your repo can't explain, asks, and turns the answers into the files your agents actually read.
Do you train on my data?
No. And the context you build is yours — one command exports all of it, anytime.
What happens if I cancel?
Your repo already holds the rendered agent files — they keep working. ch export takes everything else. No lock-in by design.
Start shaping context.
Early access is rolling out to the waitlist. Design partners get pricing locked at launch rates.