MeshTale runs exact-match and meaning-based search side by side, merges them by rank, and hands back a short, cited answer from the asking agent's own workspace.
Free tier, no card required.
INV-2207 rounding off by a centby 
Marco·Jira issue
Fix INV-2207: invoice exportby Claude Code·session
#northwind-billing: INV-2207by 
Dana·Slack threadFix INV-2207: invoice exportby Claude Code·session
Priya·Granola call
Northwind SOW v2, billing termsby 
Gemini·summaryFix INV-2207: invoice exportby Claude Codeexact #2meaning #1INV-2207 rounding off by a centby Marcoexact #1Priyameaning #2Exact-match search is best at names, ticket IDs and error strings. Meaning-based search is best at questions asked in different words. Their scores are not on the same scale, so MeshTale runs them separately and merges the two lists by position.
Your assistant calls MeshTale only when it needs context, and gets back just the cited pieces that answer the question. Your session stays fast and keeps its room for the actual work.
Northwind SOW v1by Priya·Google DriveOutdatedNorthwind SOW v2by Priya·Google DriveCurrentWhen a document or thread changes, the new version takes over and the old one stops coming back in answers. When two different memories disagree, a person is asked to decide. How conflicts work
by Claude CodeAIsession · Tueby MarcoHumanJira · MonPriyaHumanGranola · 11 JunEach piece comes back with where it came from and who wrote it, a person or an AI tool, so anyone can check the answer against the original.
Recall searches only the workspace the agent is connected to. Other clients are not filtered out afterwards. They are never searched.
Each recall comes back within a size budget: the most relevant cited pieces first, never whole documents pasted into your chat.
Ask the same thing again and the answer comes straight from cache, still scoped to the same workspace and still logged.
Not everyone on the team runs a coding agent. Tela is the chat inside MeshTale: ask a question about a workspace in plain words and it shows which sources it searched, then answers with numbered citations.
G 
Invoice totals were off by a cent because the export rounded per invoice2. It now rounds per line, which matches Northwind's ledger3. The old rounding is kept behind a flag for March reports1.
by Claude Code·sessionTueby Marco·Jira issueMonPriya·Granola call11 JunThe same search, the same rules and the same log sit behind the chat, your agents and your own code.
Two tools over MCP. The agent asks when it needs context and gets cited pieces back.
A REST API with workspace-scoped keys, so a script can only ever search the workspace its key belongs to.
Tela, for anyone on the team who wants an answer without opening a terminal.
Exact-match search looks for the words you typed, which is what you want for names, ticket IDs and error messages. Meaning-based search looks for the same idea in different words, so "why are totals off by a cent" can find a note about rounding.
The two searches score results on different scales, so adding the scores would let one method crowd out the other. Merging by position keeps each list's best results in play, and a piece both searches agree on rises to the top.
When a source is updated, the newer version replaces the older one and the old version stops coming back in answers. If two different memories disagree, MeshTale flags it so a person can decide which is accurate.
No. An agent connection or API key is bound to one workspace, and recall searches only that workspace. Each recall is logged with the agent, the workspace and what came back.
No. Each answer is sized to a budget, with the most relevant cited pieces first and the rest left out and counted. Repeat questions come back from cache.
Connect a source, then ask your first question in Tela or from your agent.
Free tier, no card required.