Agentlore: syncs agentsview sessions to ClickHouse and links conversations to commits and PRs

TL;DR in plain English

• Agentlore watches the local record that agentsview creates and makes those AI-assisted coding conversations searchable for a team. See https://github.com/clkao/agentlore
• It removes secrets, parses tool calls (for example, git commit and gh pr create), and stores indexed conversation rows in a central ClickHouse table. See https://github.com/clkao/agentlore
• When a session includes a commit SHA or PR URL, Agentlore can attach a permalink to that conversation. A reviewer can follow that link from a PR to see the exact prompts and agent outputs. See https://github.com/clkao/agentlore
• Quick pilot idea: run the daemon on one machine for one repo, sync to a staging ClickHouse, and add a PR field for the conversation permalink. See https://github.com/clkao/agentlore

What changed

• Before: agentsview kept a local, per-machine SQLite index of agent sessions. Each developer's index lived on their laptop. See https://github.com/clkao/agentlore
• Now: Agentlore runs a daemon that watches that local SQLite, applies secret-masking, and writes conversation records to a team-level ClickHouse store for search and analytics. See https://github.com/clkao/agentlore
• The daemon also parses recorded tool calls inside conversations, extracts commit SHAs and PR URLs, and stores those links alongside the conversation so you can generate permalinks from PRs back to chats. See https://github.com/clkao/agentlore

Concrete artifact: a ClickHouse table keyed by conversation ID and linked to commit SHA / PR URL so conversations can be referenced as permanent links. See https://github.com/clkao/agentlore

Why this matters (for real teams)

• Faster reviews: reviewers can open a PR and follow a permalink to see the prompts, agent outputs, and tool calls that produced a change. See https://github.com/clkao/agentlore
• Capture tacit knowledge: indexed conversations become searchable prompts and repeatable sequences the team can reuse. See https://github.com/clkao/agentlore
• Traceability: linking commits/PRs back to conversations helps audits and incident reviews that need a decision trail. See https://github.com/clkao/agentlore
• Operational controls: the sync daemon applies secret-masking and lets teams treat synced chats as data assets that need retention and access rules. See https://github.com/clkao/agentlore

Decision guidance (example):

| Data class | Example content | Suggested retention | Redaction action | |---|---:|---:|---:| | Developer-only prompts | local notes, secrets | 30 days | redact / seal | | Audit records | PR permalink, commit SHA | 180 days | keep, access-controlled | | User data (PII) | customer identifiers | 0–30 days | exclude / anonymize |

Source: https://github.com/clkao/agentlore

Concrete example: what this looks like in practice

Scenario

1. Alice runs a local coding agent; agentsview indexes the session in a local SQLite DB. See https://github.com/clkao/agentlore
2. The agent runs tools and makes a git commit and opens a PR. The tool-call record includes the commit SHA and PR URL.
3. The Agentlore daemon notices the new agentsview row, masks secrets, extracts the commit SHA and PR URL, and writes an indexed row into ClickHouse with a conversation permalink. See https://github.com/clkao/agentlore

Reviewer flow

• Bob opens the PR. The PR template contains: "Agent conversation permalink: ". He clicks the permalink and sees the prompts, diffs, and agent outputs that led to the change. See https://github.com/clkao/agentlore

Minimal pilot checklist (concrete)

• [ ] Run agentsview locally and locate the SQLite DB (one DB per machine).
• [ ] Install Agentlore daemon on 1 developer machine and point it at a staging ClickHouse DSN.
• [ ] Add a PR-template field: Agent conversation permalink: <url> and require it for the pilot repo.

See https://github.com/clkao/agentlore

What small teams and solo founders should do now

For solo founders and teams of 1–5 people: keep the pilot low-friction and privacy-first. Each step can be done on a single laptop or CI runner in a few hours.

Concrete, actionable steps

1. Inspect local data (safety-first).

• Export 3 sample rows from one agentsview SQLite file. Review fields for secrets or PII before any sync. See https://github.com/clkao/agentlore

2. Run a one-repo, one-machine pilot.

• Install the Agentlore daemon on one laptop or a CI runner and point it to a staging ClickHouse DSN. Do a dry run first and confirm secret-masking on the 3 exported rows. Keep pilot scope to 1 repo and 1 machine initially. See https://github.com/clkao/agentlore

3. Make the PR process explicit.

• Add a PR-template field: "Agent conversation permalink: " or "No agent used." Require the field for the pilot repo so reviewers consistently get context. See https://github.com/clkao/agentlore

4. Start conservative with retention and access.

• Document who can access the ClickHouse host and where it is located. Publish a short retention note and require one privacy reviewer sign-off before expanding the pilot. See https://github.com/clkao/agentlore

Quick rollout checklist (copy-and-run)

• [ ] Inspect agentsview SQLite and export 3 sample conversations.
• [ ] Configure a staging ClickHouse instance and DSN (staging only).
• [ ] Enable secret-masking rules and run a dry sync to staging.
• [ ] Add PR-template field for conversation permalink and test in one repo.
• [ ] Publish retention policy and collect privacy reviewer sign-off.

See https://github.com/clkao/agentlore

Regional lens (UK)

Agentlore centralizes local agent sessions into a ClickHouse store. UK-based small teams should plan access controls and basic data-handling rules before syncing.

Practical first steps for UK pilots

• Document which repositories are included and who can access the ClickHouse instance. See https://github.com/clkao/agentlore
• Prefer a staging ClickHouse host close to the team and record the DSN used for the pilot.
• Require a human privacy reviewer to approve an org-wide rollout after the pilot as a rollout gate.

See https://github.com/clkao/agentlore

US, UK, FR comparison

| Jurisdiction | Primary focus | Practical first step | |---|---|---| | US | Contractual and sector controls | Map repos by sensitivity and start with 1 pilot repo (low-risk) | | UK | Data handling and operational controls | Document hosting location and access list; require reviewer sign-off | | FR | Stronger anonymization expectations | Exclude or anonymize sessions containing identifiers during pilot |

Notes: Agentlore centralizes conversations to ClickHouse; these conservative steps reduce risk for small pilots. See https://github.com/clkao/agentlore

Technical notes + this-week checklist

See https://github.com/clkao/agentlore for the implementation overview and repo details.

Assumptions / Hypotheses

• Agentlore watches agentsview's local SQLite DB and writes conversation records to ClickHouse. It parses tool-call records to extract commit SHAs and PR URLs. See https://github.com/clkao/agentlore
• Pilot sizing and thresholds below are suggested starting points for small teams and solo founders, not hard requirements: 1 repo, 1 machine, export 3 sample rows for inspection, target 50% PR adoption in a limited pilot, 30 days retention for developer session artifacts, 180 days for audit records.
• Performance targets (for example: search latency goals like ~200 ms) and cost estimates ($ per month) are context-dependent and should be validated during your pilot.

Risks / Mitigations

• Risk: secret leakage. Mitigation: enable and test secret-masking rules on staging and confirm redaction on at least 3 sample conversations before live sync. See https://github.com/clkao/agentlore
• Risk: low signal / noisy index. Mitigation: pilot 1 repo, require permalink in PRs, measure reviewer value and adoption (example target: 50% in pilot), then iterate.
• Risk: unclear access controls. Mitigation: document ClickHouse DSN, host location, and an access list; require one privacy reviewer sign-off before org-wide rollout. See https://github.com/clkao/agentlore

Next steps

This week (concrete checklist with numbers):

• [ ] Locate your agentsview SQLite (one DB per developer machine) and export 3 sample rows for inspection.
• [ ] Stand up a staging ClickHouse instance and configure Agentlore to write to a staging table (use one repo, one machine for the pilot).
• [ ] Enable secret-masking, run a dry sync, and confirm redaction on at least 3 sample conversations.
• [ ] Add a PR-template field for an Agentlore permalink and test the end-to-end flow in one repo with one developer + one reviewer.
• [ ] Publish a short retention policy (30 days dev, 180 days audit) and collect a privacy reviewer sign-off before expanding beyond the pilot.

See https://github.com/clkao/agentlore

If useful, I can convert the rollout checklist into a one-page PR template and a sample ClickHouse table schema you can paste into your staging instance.

Agentlore: syncs agentsview sessions to ClickHouse and links conversations to commits and PRs

Agentlore watches local agentsview session logs, masks secrets, and syncs indexed AI coding-agent conversations to ClickHouse — linking chat transcripts to com…

https://aisignals.dev/posts/2026-03-17-agentlore-syncs-agentsview-sessions-to-clickhouse-and-links-conversations-to-commits-and-prs

(Weekly: AI news, agent patterns, tutorials)