Almanac's Company-Context Agent: How YC S26 Wires Internal Knowledge into Every LLM Call

Description and use

Almanac builds a live wiki that aggregates events from Slack, Gmail, and GitHub, enforces row‑level access control, and manages a token budget so LLM agents receive up‑to‑date context without leaks or prompt overflow.

Problem

Multi‑agent financial systems lose context between calls, hallucinate org structure, or repeat questions because company policies and documents are static, un‑secured, and token‑heavy.

Outcome

Almanac delivers up‑to‑date, permission‑aware company context to LLM agents with a predictable token budget, reducing hallucinations and improving task accuracy.

Workflow

• 1. Connect tools via OAuth and stream events to a Postgres queue.
• 2. Wiki compiler consumes events, groups them into pages (customer, project) and writes to Postgres.
• 3. Context selector queries the wiki using keyword and semantic search, filtered by user ACL, returning top pages.
• 4. Agent reads top pages, summarizes if >3,000 tokens, and proceeds with task or asks clarifying questions.
• 5. Token budget is recalculated if additional pages are requested.
• 6. Event updates (edits, recalls, closures) trigger real‑time re‑compilation to keep context fresh.

Evidence

Almanac’s approach is to maintain a live wiki that compiles activity from connected tools (Slack, Gmail, Granola notes, GitHub issues) and updates pages in real time.

Source

The system uses a tiered budget (values estimated based on typical LLM constraints and observed behavior from the product demo).

Source

The wiki compiler re-processes the thread and updates the relevant page when a Slack message is edited.

Source