meetmind 0.1.5

Turn meeting transcripts into a temporal context graph for coding agents.

meetmind

Turn meetings into a living context graph your coding agent can build from.

Every meeting produces decisions, requirements, and constraints — then they're lost in a transcript nobody re-reads. meetmind extracts them into a temporal knowledge graph that stays current across meetings: when a later call reverses a decision, the old one is marked superseded and linked to its replacement. A coding agent reads the graph as the source of truth, and the git diff of each meeting tells it exactly what changed.

recording / paste / doc ─▶ extract (Groq) ─▶ merge into graph.json (git) ─▶ coding agent
        (Meetily, free)         └── decisions · requirements · constraints · supersessions ──┘

Features

• Temporal graph, not notes — every node has a status (active / superseded / deprecated) and a supersedes link, so the graph always shows current truth and what changed.
• Many inputs — a Meetily recording, an audio file, a pasted summary, or a transcript file.
• record and walk away — arm a meeting, and meetmind auto-ingests the moment the call ends.
• Multi-project routing — one meeting that touches several projects fans out into separate per-project graphs, with cross-project dependency edges preserved.
• Agent-ready export — emit a clean, active-only spec (Markdown or JSON) grouped by feature.
• Git-native history — each meeting is a commit; meetmind delta is the change set to act on.
• Slack bot — drop a recording in a channel, get the graph back as an image.
• Works out of the box — a shared Groq key ships with the package; set your own for heavier use.

Installation

pip install meetmind             # core CLI
pip install "meetmind[all]"      # + Slack bot, graph images, PDF/DOCX ingest, embeddings

[!TIP] meetmind runs with zero setup thanks to a bundled shared Groq key. For anything beyond light use, get a free key at console.groq.com and set GROQ_API_KEY to avoid the shared rate limit.

Optional extras: [bot] (Slack), [viz] (PNG graphs — also needs the dot binary), [docs] (PDF/DOCX), [embeddings] (semantic entity resolution).

Quick start

Record a live meeting

Capture with Meetily (free, local, private), and let meetmind watch for the finished transcript:

meetmind record --link https://meet.google.com/xyz

meetmind arms, waits for you to start recording in Meetily, detects when the call ends (the transcript stops growing), then ingests and commits automatically. Press Ctrl+C to finalize early. Add --route to split a multi-project meeting across graphs.

Ingest anything else

meetmind ingest --meetily                       # latest Meetily meeting
meetmind ingest --paste "we'll use GraphQL and Postgres" --meeting-id m1
meetmind ingest transcript.txt --meeting-id 2026-07-12-kickoff
meetmind ingest --audio call.mp3 --meeting-id m1        # Groq Whisper fallback
meetmind ingest --new                           # every Meetily meeting not yet in the graph

See and use the graph

meetmind show                    # current active nodes
meetmind delta                   # git diff of the last meeting's changes
meetmind timeline                # meeting-by-meeting evolution
meetmind viz --out graph.png     # render the graph
meetmind export --format md      # clean active-only spec for a coding agent
meetmind validate                # integrity check (dangling edges, supersede cycles)

How the graph works

When a later meeting revises a decision, meetmind decides new vs. duplicate vs. change using, in order: explicit supersession from the extractor, optional embedding similarity, then string/concept matching. A changed decision doesn't overwrite the old one — the old node flips to superseded and a SUPERSEDES edge points from the new node to it.

{
  "id": "decision-api-style",
  "type": "Decision",
  "title": "API uses GraphQL",
  "content": "Switch the public API from REST to GraphQL.",
  "status": "active",
  "source_meeting": "2026-07-19-review",
  "supersedes": ["decision-api-rest"],
  "confidence": 0.9
}

Node types: Feature, Requirement, Decision, Constraint, Component, Interface, OpenQuestion, ActionItem. Edges: IMPLEMENTS, DEPENDS_ON, BLOCKS, SUPERSEDES, DECIDED_IN, RELATES_TO.

For the coding agent

Point your agent at graph.json (or meetmind export) and treat it as follows:

• Current requirements = every active node. Obey these.
• superseded / deprecated nodes are not current truth — use them to know what to replace.
• OpenQuestion nodes are undecided — flag them, don't assume an answer.
• What changed this meeting = meetmind delta. A node flipping to superseded plus a new active node linked by SUPERSEDES means "replace the old implementation with the new."

Multi-project meetings

When one meeting spans several projects, --route labels each node with a project and fans them into separate context-graphs/<project>.json files — each with its own history, supersession, and export spec.

meetmind ingest big-meeting.txt --meeting-id m1 --route
meetmind export --project payments

Known projects are discovered from existing graph files so labels stay consistent; a new project name auto-creates its graph. Cross-project dependencies are preserved as namespaced references (e.g. Billing DEPENDS_ON auth:decision-api-graphql) and surfaced in the dependent project's export.

Slack bot

Let anyone drop a meeting recording into a channel and get the graph back as an image.

pip install "meetmind[bot]"
export SLACK_BOT_TOKEN=xoxb-...      # bot token
export SLACK_APP_TOKEN=xapp-...      # app-level token, scope connections:write
export GROQ_API_KEY=...
meetmind-bot

Mentions: @bot graph (post the image), @bot show (list active nodes), @bot delta (git diff), @bot note <text> (ingest pasted text). Uploaded audio and documents are ingested automatically. Set CG_ROUTE_PROJECTS=1 to route each meeting by project instead of by channel.

[!NOTE] The bot uses Socket Mode — no public URL needed. Enable Socket Mode, subscribe to the message.channels and app_mention events, and invite the bot to the channel.

Configuration

Variable	Purpose
GROQ_API_KEY	Your Groq key (overrides the bundled shared one).
CG_CONFIDENCE_THRESHOLD	Nodes below this confidence become proposed for review (default 0).
CG_ROUTE_PROJECTS	1 = bot routes meetings by project instead of by channel.
GEMINI_API_KEY	Enables Gemini embeddings for semantic entity resolution.
SLACK_BOT_TOKEN / SLACK_APP_TOKEN	Slack bot credentials.

Any variable can also live in a .env file in your working directory.

How it captures meetings

meetmind reads Meetily's local SQLite database read-only — Meetily records system + mic audio and transcribes it locally for free, so meetings never leave your machine. A previously recorded call works too: import it into Meetily, or point meetmind straight at the file with --audio (Groq Whisper transcribes it).

[!IMPORTANT] record watches Meetily — you start the recording once in Meetily, and everything after (detection, ingest, commit) is automatic. Meetily exposes no API to start recording programmatically.