They fail because they're new.
Every confident, wrong line of AI code costs you tokens, review cycles, and trust. Decispher captures the decisions, conventions, and constraints from the conversations your team is already having in Slack, GitHub, and Jira, then serves them back to every human and every agent that needs them. Automatically.
[NEW](#mcp)
**Chat Clipboard**· transfer mid-task context across sessions, machines & AI tools[NEW](#branch-story)
Branch Story· every branch has a story, never start a session cold again## Your team has the answers.
Your new hires and AI agents don't.
Every engineering team has two kinds of knowledge. The explicit kind lives in GitHub: code, READMEs, tickets. The tacit kind lives in people's heads: why the architecture is that way, what was tried and failed, which invariant must never break. AI agents only see the first one.
Three jobs. One system of record. #
Decispher reads the conversations your team already has, structures the durable signal into seven canonical context units, and serves them back to every consumer that needs them: humans in their IDE and dashboard, AI agents over MCP.
From the work, not on top of it.
Bots listen in Slack channels. Webhooks watch GitHub PRs and reviews. The Jira app reads ticket threads and feeds context back when an agent picks the ticket up. No new tool for engineers, no manual documentation step. Decisions are extracted from the conversations, tickets, and PRs your team is already having.
The same decision, said three times, becomes one.
A five-step LLM pipeline classifies every message, extracts the structured why, and fuses fragments across channels. Same idea in Slack and PR description? Merged. Multi-source agreement boosts confidence. Conflicts are flagged.
Every consumer, the right amount, on demand.
Humans get search, “Ask Decispher,” and PR comments. Agents get MCP tools and 9 native instruction files. And when code contradicts a decision on record, the PR gets flagged before a reviewer ever opens the diff.
Read where the work happens. Change nothing. #
Ninety percent of engineering decisions are made in conversation and never documented. If you can't capture there, you can't capture at all. Decispher plugs in as a Slack bot, a GitHub webhook, and a Jira app. Your team writes the same messages they always have; we do the rest.
The same decision, said three times, #
becomes one record.
A constraint mentioned in Slack, repeated in a PR description, and confirmed in a design doc should not become three rules. Decispher's Fusion engine embeds every fragment, finds the ones that mean the same thing, and merges them into a single, multi-sourced unit. Cross-channel agreement raises confidence. Disagreement raises a flag.
A database of decisions is just a list. #
A graph is a memory.
Decispher persists the relationships between context units, not just the units themselves. Every new decision is classified against your existing graph: does it confirm, extend, contradict, or supersede something we already know? The result is a queryable map of how every choice your team has made depends on every other one.
CONFIRMS
EXTENDS
CONTRADICTS
SUPERSEDES
SAME_TOPIC
DERIVED_FROM
Commit the rules. Serve the rest live. #
Dumping every decision into CLAUDE.md
wastes tokens on every call. Dumping nothing makes the agent guess. Decispher does neither. A tiered .decispher/
map lives in your repo with a tiny always-loaded spine, and the deep detail is served on demand over MCP. Agents discover, narrow, and deep fetch, paying only for the tokens they actually use.
Generated by Decispher, opened as a PR, version-controlled in your repo. Every agent loads it for free on every session. No fetch, no cost.
Copilot · Windsurf
The deep detail stays in Decispher and is served over the Model Context Protocol. Three tools, one chain: discover what exists, narrow to a topic, deep-fetch the unit. Every call is receipted.
list_topics
freediscover the topic namespaceget_context_for_topic
1 creditspine inline plus IDs of the restget_decision
1 creditfull body for one decision**~500 tokens** they need, instead of the
~8,000 they'd guess at. Nothing dropped, just deferred.
Humans ask in plain English. Agents query the same memory over MCP. #
Every “why did we…” gets an instant, cited answer, served in dashboard chat, a Slack command, or a GitHub PR comment. The same memory is exposed to AI agents through 12 MCP tools, so every coding tool your team uses reads from one source. Every answer cites the units it came from. No source, no claim.
decispher.check_intent
pre-flight check before any code change. Returns BLOCKED, WARN, or CLEARdecispher.search_decisions
semantic search across all 7 context types. Filter by type or search everythingdecispher.get_constraints
fetch all active architectural constraints the agent must not violatedecispher.check_conventions
retrieve all active coding conventions before writing codedecispher.get_context_for_file
symbol-graph plus embedding retrieval for one file or up to 10 at oncedecispher.ask_knowledge_base
AI-synthesised answer with cited sources from the knowledge basedecispher.list_topics
discover the project's topic namespace. Free, no credits consumeddecispher.get_context_for_topic
curated context cluster for a topic: spine units inline, expansion IDs returneddecispher.get_decision
full body for any context unit by ID: rationale, alternatives, affected filesdecispher.capture_decision
write new knowledge back into the team's knowledge base from inside an agentdecispher.copy_chat
portable conversation snapshot. Server returns a clipKey that works across machines and AI tools (7-day TTL)decispher.paste_chat
restore a snapshot by clipKey. Becomes the new session's working memory · free+6 freeTurn on
Branch Story and agents get six more tools for branch-keyed memory:store_read
, store_write
, and four more. All free.Learn more## Every branch has a story. Don't let it evaporate.
An AI agent rebuilds its understanding of a branch from scratch every session. Yesterday's reasoning, the constraint you found the hard way, the approach that didn't work: all gone by morning. Branch Story keeps it. One command attaches Decispher to Claude Code or Cursor, then every session opens with a briefing of what's already known and quietly records what's new.
The Branch Store, Story Mode, Save Context, and “why is this line here” lookups, right inside your editor. Works in VS Code, Cursor, Windsurf, and other VS Code forks via Open VSX.
Plug into the stack your team already uses. #
Capture from where the work happens, serve to every tool that writes code. No new habits for engineers, no migration project. Everything below is live today.
Every PR makes Decispher smarter than yesterday. #
Decispher isn't a one-time integration. It's a loop. Decisions feed the Context Map; the map writes the committed rules; agents follow them; Lens watches what actually fires; the score re-ranks every rule. Every cycle through the loop, the context gets denser and the agents get more accurate. The longer it runs, the harder it is to leave.
Compound interest, on engineering memory.
Every Lens scan makes the Context Map smarter. Every regeneration makes the next scan more relevant. Competitors building the same surface area without the full loop catch up to last month's Decispher.
Receipted, not estimated. #
Every claim Decispher makes ships with a number you can verify on your dashboard. The Context Health Score is the metric your CTO will check on Monday morning.
~40%
2.9×
9
0
0
Stop paying Claude to rediscover #
what your team already figured out.
Two-week onboarding. Decispher reads your Slack, GitHub, and Jira history once, builds your initial Context Map, and ships a PR with the .decispher/
directory. Run npx decispher init
and Branch Story starts briefing every session. You approve. Your agents wake up smarter the next morning.