How it works
Your whole team's AI, on the same page
Your team already works with AI. BaseThread is the shared context underneath it: every teammate's AI logs what it does and what you decide, we harmonize it in the cloud, and push it back so everyone's AI stays current, automatically.
Witness log harmonize push back
The problem
Every AI starts from zero
Your AI re-learns the same context every session. A teammate's AI never sees what yours decided. Context lives in one person's chat history and dies there. The bigger the team, the worse it gets.
Objection
Wait, doesn't my AI already remember?
Your AI's built-in memory is personal, and it's trapped in one app. Claude remembers your chats. ChatGPT remembers your projects. Cursor reads your repo's rules. None of them share what your teammate's AI learned, and none of them work across the other tools your team uses. BaseThread is the shared layer that does.
How it works
Your AI does the work. BaseThread remembers it, for everyone.
Reads before it works
Your AI pulls the latest team context through MCP, so it starts every task current. Desktop clients go through the BaseThread Mac app's local MCP server; cloud clients like ChatGPT go straight to mcp.basethread.ai.
Logs as it goes
It records activity (what happened), decisions (what you settled, and why), and tasks (what's next, and who owns it) into the shared ledger. No manual note-taking.
Harmonized and shared back
We reconcile everyone's streams in the cloud and push the canonical version back to every teammate's AI.
Activity Decisionsand Tasksflow through the same ledger.
The third stream
Not just memory. It coordinates the work.
Activity and decisions look back; tasks look forward. Your AI creates, assigns, and tracks tasks across tools, not just remembers them.
Create from any tool
“Add a task to follow up with Globex after the demo, due Friday.” Created in place, in whatever AI you’re already in.
Assign to a teammate
Delegate the pricing-copy sign-off to PMM. It lands in their list, and their AI sees it at the start of the next session.
Track to done
Ask what’s on your plate, what you delegated, and what’s still open. Status is read from the shared task stream, not a guess.
The five levels
Context, shaped like your company
Company → Products → Teams → Projects → Me. Your AI pulls exactly what's relevant to the task in front of it, not a wall of everything.
The context graph
One question, the exact right context
Every node, every link, queryable. When a question lands, the graph routes through the few nodes that matter, so your AI answers with citations instead of a generic guess.
A question lands
“What did we decide about pricing, and why?”
Asked in Cursor by Maya
The graph routes
- Decision: Move to simple per-user pricing
- Activity: Q1 pricing review meeting
- Activity: Cohort pricing feedback
- Project: Pricing v2 · Products layer
Pulled from exactly the nodes that matter. Not the whole graph.
Your AI answers, grounded
“You moved to simple per-user pricing in the Q1 review. Usage caps and per-action billing were parked. Rationale: most accounts are small teams.”
Citations: decision:7f2a, activity:8c2a, activity:91f3
Works with the AI you already use
Two ways your AI reaches your team's context
The BaseThread Mac app spins up a local MCP server on your machine for desktop AI. Cloud-bound AI like ChatGPT connects to our remote MCP at mcp.basethread.ai. Both paths read and write the same team context, no new app to learn.
Path 1 · Local MCP
The Mac app spins it up
The Mac app spins up a local MCP server on your machine. Any MCP-capable AI on the same machine connects to it, reads your team's context, and writes activity, decisions, and tasks back. Fast and offline-capable.
your-mac → local MCP → BaseThread
Path 2 · Remote MCP
Hosted at mcp.basethread.ai
Our hosted MCP endpoint at mcp.basethread.ai. No install. AI that runs in the cloud or a browser connects directly. The right call for clients that can't use a local server, like ChatGPT or hosted agents.
cloud AI → mcp.basethread.ai → BaseThread
Both paths reach the same BaseThread cloud and the same harmonized team ledger. Pick whichever fits the AI you're using; switch back and forth without losing context.
Harmonization
Everyone on the same page, without anyone managing it
BaseThread merges duplicate activity beats, reconciles conflicting decisions, links related entries across streams, and retires the ones that got superseded, so your team's shared context stays clean and current on its own.
Raw streams
- Sarah's Claude CodeActivityShipped the magic-link auth flowkept
- Maya's CursorDecisionMove to simple per-user pricingkept
- Raj's ClaudeActivityShipped magic-link auth (per Sarah's PR)duplicate
- Mia's ChatGPTDecisionPer-action pricing parkedsuperseded
- Priya's CursorTaskPricing-copy sign-off, due Frikept
Canonical ledger
- Shipped the magic-link auth flowActivityLogged by Sarah's Claude Code. Duplicate from Raj's Claude merged.
- Move to simple per-user pricingDecisionSupersedes "Per-action pricing parked." Linked.
- Pricing-copy sign-off, due FriTaskAssigned to PMM. Open. Linked to the pricing decision.
Your context, protected
Encrypted in transit and at rest, role-based access at every level, and never used to train AI models. You own your data.